Patent Application
20090268724
The present invention relates to the fields of integrated circuits, networking, systems and processes for packet communications, and especially communication of real time information such as voice, audio, images, video and other real time information over packet.
The Internet has long been usable for Internet file transfers and e-mail by packet switched communication. A different technology called circuit switched communication is used in the PSTN (public switched telephone network) wherein a circuit is dedicated to each phone call regardless of whether the circuit is being communicated over in silent periods. Packet switched networks do not dedicate a channel, thereby sharing a pipe or channel among many communications and their users. Packets may vary in their length, and have a header for source information, destination information, number of bits in the packet, how many items, priority information, and security information.
A packet of data often traverses several nodes as it goes across the network in “hops.” In a stream of data, the packets representative thereof may, and often do, take different paths through the network to get the destination. The packets arrive out of order sometimes. The packets are not only merely delayed relative to the source, but also have delay jitter. Delay jitter is variability in packet delay, or variation in timing of packets relative to each other due to buffering within nodes in the same routing path, and differing delays and/or numbers of hops in different routing paths. Packets may even be actually lost and never reach their destination. Delay jitter is a packet-to-packet concept for the present purposes, and jitter of bits within a given packet is a less emphasized subject herein.
Voice over Packet (VOP) and Voice over Internet Protocol (VoIP) are sensitive to delay jitter to an extent qualitatively more important than for text data files for example. Delay jitter produces interruptions, clicks, pops, hisses and blurring of the sound and/or images as perceived by the user, unless the delay jitter problem can be ameliorated or obviated. Packets that are not literally lost, but are substantially delayed when received, may have to be discarded at the destination nonetheless because they have lost their usefulness at the receiving end. Thus, packets that are discarded, as well as those that are literally lost, are all called “lost packets” herein except where a more specific distinction is made explicit or is plain from the context.
The user can rarely tolerate as much as half a second (500 milliseconds) of delay, and even then may avoid using VOP if its quality is perceptibly inferior to other readily available and albeit more expensive transmission alternatives. Such avoidance may occur with delays of 250 milliseconds or even less, while Internet phone technology hitherto may have suffered from end-to-end delays of as much as 600 milliseconds or more.
Hitherto, one approach has stored the arriving packets in a buffer, but if the buffer is too short, packets are lost. If the buffer is too long, it contributes to delay.
If the network is very congested, and the packet is routed by a large number of hops, the ratio of lost packets to sent packets in a given time window interval can rise not just to 5-10% but even to 25% or more, and the real-time communication becomes degraded. VOP quality requires low lost packet ratio measured in a relatively short time window interval (length of oral utterance for instance, with each packet representing a compressed few centiseconds of speech). By contrast, text file reception can reorder packets during a relatively much longer window interval of reception of text and readying it for printing, viewing, editing, or other use. Voice can be multiplexed along with other data on a packet network inexpensively over long distances and internationally, at low expense compared with circuit-switched PSTN charges.
A Transport Control Protocol (TCP) sometimes used in connection with the IP (Internet Protocol) can provide for packet tags, detection of lost and out-of-order packets by examination of the packet tags and retransmission of the lost packets from the source. TCP is useful for maintaining transmission quality of e-mail and other non-real-time data. However, the delay inherent in the request-for-retransmission process currently may reduce the usefulness of TCP and other ARQ (automatic retransmission request) approaches as a means of enhancing VOP communications.
RTP (Real Time Transport Protocol) and RTCP (RTP Control Protocol) add time stamps and sequence numbers to the packets, augmenting the operations of the network protocol such as IP. However, these do not provide QoS (Quality of Service) control.
For real-time communication some solution to the problem of packet loss is imperative, and the packet loss problem is exacerbated in heavily-loaded packet networks. Also, even a lightly-loaded packet network with a packet loss ratio of 0.1% perhaps, still requires some mechanism to deal with the circumstances of lost packets.
A conventional speech compression algorithm has a portion that samples, digitizes and buffers speech in a frame buffer in frame intervals (e.g. 20 milliseconds), or frames, and another portion that compresses the sampled digitized speech from one of the frames while more speech is being added to the buffer. If the speech is sampled at 8 kiloHertz, then each 20 millisecond example frame has 160 analog speech samples (8×20). If an 8-bit analog to digital converter (ADC) is used, then 1280 bits (160×8) result as the digitized form of the sampled speech in that 20 millisecond frame. Next the compression algorithm converts the 1280 bits to fewer bits carrying the same or almost the same speech information. Suppose the algorithm provides 8:1 compression. Then 1280/8 bits, or 160 bits of compressed or coded speech result from compression. The compressed speech is then put in the format of a packet, thus called packetized, by a packetizer process.
For every frame of compressed speech in a packet, loss of that packet means loss of each frame in that packet. There then arises the problem how to create 160 bits or more of lost compressed speech. One known approach simply repeats the most recent previous frame that is available at the receiving destination. Another known approach fills the output frame with silence (zeroes). Reduction of packet loss and packet loss handling strategy are very important challenges in advancing VOP technology.
In one form of the invention, a process of sending packets of real-time information at a sender includes steps of initially generating at the sender the packets of real-time information with a source rate greater than zero kilobits per second, and a time or path or combined time/path diversity rate, the amount of diversity initially being at least zero kilobits per second. The process sends the packets, thereby resulting in a quality of service QoS, and optionally obtains at the sender a measure of the QoS. Another step compares the QoS with a threshold of acceptability, and when the QoS is on an unacceptable side of said threshold increases the diversity rate and sends not only additional ones of the packets of real-time information but also sends diversity packets at the diversity rate as increased. Also, rate/diversity adaptation decision may be performed at receiver.
Increasing the diversity rate while either reducing or keeping unchanged the overall transmission rate is an important new improvement in even solely-time-diversity embodiments.
Further forms of the invention involve new criteria for initiating adaptation transitions, and new types of transitions including number of packets-per-second transitions, diversity transitions, source rate transitions and mixtures thereof.
In another form of the invention a single-chip integrated circuit includes a processor circuit, and a source rate/diversity control. Here again, the diversity is contemplated to be time diversity, path diversity and combined time/path diversity in various embodiments.
Other forms of the invention encompass other processes, improved packets and packet ensembles, integrated circuits, chipsets, computer add-in cards, information storage articles, systems, computers, gateways, routers, cellular telephone handsets, wireless base stations, appliances, and packet networks, and other forms as claimed.
Various embodiments provide adaptive, robust VoIP/VOP/media over packet (including real time signals over packet) solutions. They provide approaches to packet network improvements for incorporation into VoIP/VOP/media-over-packet IETF, TIPHON, and ITU standards. Packet loss resilience encoding and packet loss handling are improved. Adaptive delay and delay-jitter handling contribute to efficient playout and congestion detection. An adaptive delay and/or delay-jitter handling mechanism is integrated with speech, audio, video and image coders. Constrained rate/diversity adaptation processes and systems embodiments control congestion robustly.
In packet loss resilience encoding and packet loss handling, sender based diversity embodiments improve G.729 and Texas Instruments code excited linear prediction (TI-CELP) codec among other coders. The following document is hereby incorporated herein by reference for use where G.729 is referred to herein: International Telecommunication Union ITU-T G.729 (03/96) Telecommunication Standardization Sector of ITU, General Aspects of Digital Transmission Systems, Coding of Speech at 8 kbit/s Using Conjugate-Structure Algebraic-Code-Excited Linear-Prediction (CS-ACELP), ITU-T Recommendation G.729.
For example, information about packet n is sent in packets {n+k: k>0} in a packet sequence:
[P(n−1)′P(n)][P(n)′P(n+1)][P(n+1)′P(n+2)][P(n+2)′P(n+3)]
Computationally-efficient CELP based important information redundancy schemes are provided.
Computationally-efficient multiple description CELP coding is provided.
Adaptive delay and adaptive delay-jitter handling advantageously compensate delay variation in arriving packets, detect delay-spikes of delay value due to congestion, and increase playout delay and send congestion notification.
Adaptive delay and adaptive delay-jitter handling process is suitably integrated with G.729 codec and other codecs.
Combined adaptation on both source rate sij and packet network diversity rate dij, a process called rate/diversity adaptation herein, robustly controls congestion. Some embodiments herein use source rate adaptation alone, with advantageous simplicity and QoS improvement compared to approaches hitherto. Also, further embodiments use diversity adaptation alone or combined with source rate adaptation with the following advantages for real-time traffic:
To handle congestion, TCP reduces the number of packets transmitted and uses retransmission which often introduces unacceptable delay and delay-jitter for real-time traffic. In rate/diversity adaptation for real-time communication, diversity is advantageously introduced.
As is described herein, rate/diversity adaptation for robust congestion control offers features in one or another of the embodiments, such as
Important areas of improvement for VoIP/VOP technology involve minimizing delays inside computers and their software, lowering network latency, and tightening network jitter. One or more of these advantages are conferred by some of the embodiments described herein.
By adapting transmission rate and the amount of time or path or combined time/path diversity in VoIP/VOP applications, robust solutions advantageously handle network impairments and congestion, while utilizing network resources efficiently.
Improvements in VoIP/VOP processes, integrated circuits and systems utilizing path diversity are described in the coassigned patent U.S. Pat. No. 6,496,477 “Integrated Circuits, Systems, Apparatus, Packets and Processes Utilizing Path Diversity for Media Over Packet Applications,” which is incorporated herein by reference. In one category of embodiments, the skilled worker uses the circuits and methods described in the incorporated material and adds the adaptive features further described herein.
RTP/UDP/IP protocols do not offer QoS control mechanisms. Hence, VoIP applications, if they were to use RTP/UDP/IP protocols, suffer from fluctuations in network conditions and poor voice quality can result. One approach for QoS control involves source rate control, with no diversity, wherein one approach for QoS control is to adapt the source rate to the fluctuations in network conditions, per “Reducing bandwidth requirements,” Micom Whitepaper, 1998; D. Sisalem et al., “The loss-delay based adjustment algorithm: A TCP-friendly adaptation scheme,” NOSSDAV, (International Workshop on Network and Operating System Support for Digital Audio and Video), July 1998. However, this approach may not handle short-term network fluctuations well, and is complicated as VoIP/VOP applications often involve multiple links of heterogeneous characteristics. First, there is a need to locate the “bottleneck” link, and, second, all users of the bottleneck link may not reduce their transmission rate.
In time diversity, information about packet n is also transmitted in packet n+1 and sometimes in even further packets where packets having at least some information in common with each other are called dependent packets.
Path Diversity sends dependent packets over two or more paths in the network, thus increasing the probability of recovering the information that was coded to produce the dependent packets.
Combined Time/Path Diversity approach uses both processes of Time Diversity and Path Diversity in innovative ways.
“Diversity packet,” where the term is used herein sometimes means a self-contained packet with its own header and diversity information. However, the term “diversity packet” can also mean diversity bits and extra header bits put in a packet that already has a header and a payload.
Time diversity schemes provide inter-packet diversity, by including information about the nth packet in succeeding packets {n+k: k>=1}. They may employ redundancy schemes (media-specific redundancy, forward-error correction FEC) and multiple-description schemes, for instance.
In the example sequence of four packets just below, bits P(n) represent primary packets, and P(n)′ and P(n)″ each represent instances of diversity. This packet sequence has a number of diversity stages (here 3 namely P(n), P(n)′ and P(n)″) and a diversity length of 4. Diversity length is the minimum number of packets in a symbol sequence needed to define the diversity used.
P(n)P(n−1)′P(n−3)″P(n+1)P(n)′P(n−2)″P(n+2)P(n+1)′P(n−1)″P(n+3)P(n+2)′P(n)″
The diversity length is greater than the number of diversity stages because of diversity offset, which is one here. Diversity offset corresponds in this case to absence of P(n) or any primed P(n) in the third packet.
Redundancy schemes piggyback a version/function (media-specific redundancy/FEC) of nth packet to (n+k)th packet, k>=1, as shown hereinbelow. The following sequences of packets are examples of media-specific redundancy schemes:
where Pn denotes nth packet, and Pn′ and Pn″ denote versions of the nth packet. “Version” here means a dependent datum in a packet having at least some information included or encoded therein which corresponds to another packet. Thus, the juxtaposition of symbols used above signifies concatenation in the description of a given packet, and any possible sequence of concatenation in every packet is contemplated, as well as the order of concatenation literally shown for each packet.
Multiple-description (MD) schemes break the input stream into multiple descriptions, for instance, using MD quantizers [V. A. Vaishampayan et al., “Asymptotic analysis of multiple description quantizers,” IEEE Trans. On Inform. Theory, January 1998}. Here none of the descriptions have the full information intended for reception, and instead each of the descriptions has less than that full information, and the descriptions which are received (even if some be lost) then have their information combined in the decoder to obtain what information is available in them collectively. The following packet sequences symbolize examples of embodiments of process and systems
where Pn, Pn′ and Pn″ denote multiple descriptions of the nth packet.
One type of embodiment uses plural types of time diversity concurrently so that Redundancy is applied concurrently with Multiple-Description.
Among other advantageous things herein, the present application describes path diversity processes, integrated circuits and systems whereby VoIP/VOP software applications open multiple (two or more) flows between the same source and the destination. The packets in each flow traverse separate paths from packets in other flows (for at least some of the hops between the source and destination). By having multiple paths, or causing multiple network paths to be accessed, used and traversed, such path diversity processes, integrated circuits and systems reduce, as between the diverse flows, the correlation of packet loss, delay, jitter and other less than desirable metrics of performance which are ameliorated herein.
Some examples of combined time/path diversity embodiments for VOP that confer advantageously efficient bandwidth utilization are described next.
1. Combined time/path diversity. The time diversity is redundancy, multiple description, FEC forward error correction, or other suitable process.
(In embodiment 1(b), the packet stream of Path 2 is time-delayed relative to the packet stream of Path 1.)
2. Path switching diversity randomizes bursty packet losses without increasing bandwidth utilization. First create multiple connections/paths (e.g. 2 connections) between the source and the destination. Then transmit as follows:
Generally speaking, a set of n respective packets are directed into a corresponding number n of respective diverse network paths whereupon the process repeats for the next set of n packets, and so on.
3. Combined path-switching diversity/redundancy embodiments combine path-switching diversity and redundancy processes, and advantageously achieve good voice quality with efficient bandwidth utilization. In a two-path embodiment
In this approach, n respective redundancy packets are directed into a corresponding number n of respective diverse network paths whereupon the process repeats for the next set of n packets, and so on.
4. Combined path-switching diversity/multiple-descriptions embodiments combine path-switching diversity and multiple-description processes, and advantageously achieve good voice quality with efficient bandwidth utilization. In a two-path embodiment
In this approach, n respective multiple-descriptions packets are directed into a corresponding number n of respective diverse network paths whereupon the process repeats for the next set of n packets, and so on.
5. In an embodiment of incorporated coassigned patent U.S. Pat. No. 6,496,477, the original voice packet stream (P0 P1 P2) is sent in its entirety, on path 1 and on path 2.
The following Performance Table summarizes performance for a two-path system of the above four embodiments. The symbols R, H and R′ refer to source bit rate, header bit rate, and redundancy bit rate, respectively. Concealment processes such as interpolation are suitably used to recover missing parts of a media stream. In G.729, a frame erasure concealment method specified in G.729 spec is suitably used when the path diversity feature herein is not being used.
In a similar manner, the skilled worker analyzes various embodiments and selects for implementation whichever one(s) are most suitable for the particular needs at hand.
Media-specific redundancy schemes piggy-back a version of nth packet to (n+k)th packet. In VoIP/VOP heretofore a separate encoding scheme generated the redundancy version of the nth packet, or piggybacked the entire nth packet to the (n+k)th packet. Herein computationally-efficient CELP (code-excited linear prediction) based diversity embodiments for VoIP/VOP are described, such as for generating media-specific redundancy information. These are herein called “Important Information” based diversity embodiments. Some Important Information based diversity embodiments use base information, or Important Information, from CELP encoding as redundancy information, to achieve diversity. Below, two embodiments are described in more detail for G.729. These embodiments are given for two stages (primary stage plus one secondary stage), with no diversity offset. Embodiments include extensions based on more than two stages, and diversity offsets.
Embodiment 1: With no pulses in secondary stage. Using G.729, the secondary stage (redundancy stage) has these Important Parameters—LPC (Linear Predictive Coding) parameters, LTP (Longterm Prediction) lags, parity check, and adaptive and fixed codebook gains—according to the sequence
P(n)P(n−1)′P(n+1)P(n)′P(n+2)P(n+1)′P(n+3)P(n+2)′
1A. Reconstruction with single packet loss is shown in the next sequence below. The LPC parameters, LTP lags, parity check, and adaptive and fixed codebook gains are obtained from the secondary stage. The excitation reconstruction mechanism is suitably made to be the replacement excitation generation scheme described in the G.729 standard section 4.4.4 with the following modification. For lost-frames considered as nonperiodic, the adaptive codebook contribution is set to zero only if the absolute value of the adaptive codebook gain (obtained from the secondary stage) is less than 0.4, otherwise the adaptive codebook contribution is reconstructed from the adaptive codebook gain and LTP lag obtained from the secondary stage.
(where “excitation” shown above refers to reconstruction of the dashed part of the packet symbols)
1B. Reconstruction with two or more consecutive packet losses is shown in the next sequence below. Now the packet (n+2) is reconstructed as described in the paragraph 1A just above. The packet (n+1) is reconstructed by the G.729 frame erasure concealment scheme specified in the G.729 standard section 4.4, used for packet loss concealment. The steps of section 4.4 are repetition of synthesis filter parameters (4.4.1) attenuation of adaptive and fixed-codebook gains (4.4.2), attenuation of the memory of the gain predictor (4.4.3), and generation of the replacement excitation (4.4.4).
Embodiment 2: With pulses in secondary stage. Using G.729, the secondary stage (redundancy stage) has LPC parameters, LTP lags, parity check, and adaptive and fixed codebook gains, and first few or all fixed codebook pulses.
2A. In reconstruction with single packet loss, the LPC parameters, LTP lags, adaptive and fixed codebook gains, and the included pulses are obtained from the secondary stage. The remaining fixed codebook pulses are set to zero.
2B. Reconstruction with two or more consecutive packet losses reconstructs the packet (n+2) as described in the paragraph 2A just above. The packet (n+1) is reconstructed by the G.729 frame erasure concealment scheme specified in the G.729 standard section 4.4, used for packet loss concealment, when there is no diversity.
Multiple-description data partitioning based diversity embodiments are described next.
It is believed that heretofore there has been no CELP-based multiple description process. Herein are described computationally-efficient, CELP-based multiple description embodiments using multiple-description data partitioning. Parentheses are used in the next few sentences to point out certain significant combinations of information.
These embodiments send (the base or important information+a subset of fixed excitation) in one packet and (the base or important information+the complementary subset of fixed excitation) in another packet. Below, two embodiments are described in more detail for G.729. These embodiments are given for two stages, with no diversity offset. Embodiments include extensions based on more than two stages and, diversity offsets.
DEFINITION: Multiple description data partitioning: In this approach, (the base information+a subset of enhancement information) is sent in one packet, and (the base information+the complementary subset of enhancement information) is sent in another packet. Here, when only one of the packets is received at the receiver, to produce acceptable quality that packet is reconstructed. When both packets are received at the receiver, they both are combined to produce better quality.
Embodiment 3: with no pulses in the base or important information. Using G.729, the first stage has LPC parameters, LTP lags, parity check, adaptive and fixed codebook gains, and every other fixed codebook pulses. The second stage has LPC parameters, LTP lags, parity check, adaptive and fixed codebook gains, and the remaining fixed codebook pulses. See sequence below:
P(n)P(n−1)′P(n+1)P(n)′P(n+2)P(n+1)′P(n+3)P(n+2)′
3A. In reconstruction with single packet loss, for packet n and packet (n+1), only one stage is used for reconstruction, and the remaining fixed codebook pulses are set to zero (note that these pulses include the fixed codebook pulses from the lost diversity stage). See reconstruction below:
(The plus (+) sign refers to combination of information for reconstruction).
3B. Reconstruction with two or more consecutive packet losses reconstructs the packet n and the packet (n+2) as described in the paragraph 3A just above. The packet (n+1) is reconstructed by the G.729 frame erasure concealment scheme specified in the G.729 standard section 4.4, used for packet loss concealment. See reconstruction below:
Embodiment 4: with pulses in the base or important information. Using G.729, the first stage has LPC parameters, LTP lags, parity check, adaptive and fixed codebook gains, first few fixed codebook pulses, and every other fixed codebook pulses from the remaining pulses. The second stage has LPC parameters, LTP lags, parity check, adaptive and fixed codebook gains, the same first few fixed codebook pulses, and the complementary subset of pulses from the remaining fixed codebook pulses. See sequence below:
P(n)P(n−1)′P(n+1)P(n)′P(n+2)P(n+1)′P(n+3)P(n+2)′
4A. In reconstruction with single packet loss, for packet n and packet (n+1), only one stage is used for reconstruction, and the remaining fixed codebook pulses are set to zero. See reconstruction below:
4B. Reconstruction with two or more consecutive packet losses reconstructs the packet n and the packet (n+2) as described in the paragraph 4A just above. The packet (n+1) is reconstructed by the G.729 frame erasure concealment scheme specified in the G.729 standard section 4.4, used for packet loss concealment. See reconstruction below:
Further embodiments are contemplated with
diversity offset,
multiple stages, and
multiple stages and diversity offsets.
Regarding performance and delay due to diversity: If the packet delay variation is larger than the packet interval/size, the system may choose not to introduce additional delay while making use of diversity in a limited manner.
Some other embodiments augment the MD (multiple description) approach as follows. For fixed codebook search, minimize [error(full rate)+w1 error(Description 1)+w2 error(Description 2)] instead of minimizing error(full rate) alone. (The letters “w1” and “w2” symbolize weight coefficients. Description 1 and Description 2 symbolize two descriptions). In addition an interpolation filter is used for shaping/filling of excitation. Also, MD quantizers are used for LPC parameters, LTP lags, fixed codebook gain and adaptive codebook gain.
Some Important Information embodiments apply FEC to important information.
Still other embodiments combine interleaving and diversity.
Some diversity based embodiments add interpolation of parameters in addition to fixed excitation repeating, from available (past/future) frames.
In a type of constrained adaptive rate/diversity processes, integrated circuits and systems herein, these adapt source rate and the amount of time or path or time/path diversity, in accordance with network fluctuations based on some QoS level measure (e.g., overall packet loss rate due to packet loss, delay, delay-jitter, etc., but before the application or compensation with diversity). Note that QoS is an inverse function of packet loss rate—in other words, QoS goes up as packet loss rate goes down. Thus, being higher than a threshold of QoS means being less than a corresponding threshold of packet loss rate. Put yet another way, QoS is a positive quantity and packet loss rate can be thought of as a negative quantity.
Further details of some adaptation process embodiments are as follows:
When QoS level measure is lower than a given QoS threshold (e.g., overall packet loss rate before the application of diversity exceeds (>) Threshold1), increase/introduce diversity while decreasing overall transmission rate or keeping overall transmission rate substantially unchanged.
B. When QoS level measure is higher than another QoS threshold representing higher quality of service than the given QoS threshold of paragraph A (e.g., overall packet-loss rate before the application of diversity is less than (<) Threshold2 where Threshold2 is less than or equal (<=) Threshold1), increase source rate (the bit rate for packet stream Pn). Note that Threshold1 Th1 and Threshold2 Th2 are values of the packet loss rate metric, inversely related to QoS and thresholds of QoS. The method for determining new steady-state source rate depends on available network resources according to any suitable table, algorithm or method selected by the skilled worker, see examples herein. The process of increasing source rate is achieved through one or more stages or steps. Two different steps which can either be used alone, or consecutively, or concurrently, are
Note that source rate (sij) is different from “overall transmission rate.” Overall transmission rate for purposes herein denotes the sum sij+dij in a given state. Roughly speaking, overall transmission rate is the sum of the packet P0 rate plus packet P0′ rate plus rates for any other diversity packet for P0-primed.
By the use of diversity, some process embodiments handle short-term network fluctuations well, cope with VoIP/VOP applications that involve multiple links of heterogeneous characteristics, and are TCP-traffic friendly. This is because the overall transmission rate is decreased, or at least not increased, on the network in the event of low QoS level, thereby not burdening the network increasingly, as these process embodiments work to ameliorate the QoS. In this way such process embodiments improve QoS for users and are network friendly in that such process embodiments could be implemented at one node, some nodes, or all nodes without further congesting the network while improving QoS.
In one example, packet loss rate Threshold1 is selected to be three percent (3%), and packet loss rate Threshold 2 is selected to be one-half percent (0.5%). Packet size is forty (40) milliseconds, corresponding to an overhead (header rate) of 320 bits/40 msec=8 kbps for VoIP. RTCP Transmission Interval is set to five (5) seconds, and the fraction lost (or packet-loss rate) is computed during last five (5) seconds in a latter part of the RTCP Transmission Interval. (Use of RTP and RTCP is described further later hereinbelow.) Source rate selections s11, s12, s21, s22, s31, s32 are established at 16.0, 11.2, 11.2, 8.0, 8.0 and 5.7 kilobits/sec respectively. Diversity selections d11, dl2, d21, d22, d31, d32 are established at 0.0, 4.8, 0.0, 3.2, 0.0 and 2.3 kilobits/sec respectively. All the foregoing values are, of course, offered illustratively and not in any limiting sense.
QoS level measure computation is temporally localized at the suggested 5 seconds in order to avoid smoothing of network effects. The example adaptation mechanism uses a high threshold (Threshold1) when QoS level decreases, and uses a low threshold (Threshold2) when QoS level improves. This approach advantageously addresses a scenario of possible oscillation between rate/diversity states, but where this scenario is not applicable or is addressed by other means such as delay processes or otherwise, then some embodiments can also use equal thresholds or any choice of thresholds that confers satisfactory adaptation.
Some adaptation embodiments take into account packet loss, high delay, and delay-jitter, such as in the QoS level measure process. In one type of process embodiment, overall packet-loss rate due to loss, delay and delay-jitter (and before the application of diversity) is used as QoS level measure.
Two specific embodiments or realizations of a rate/diversity adaptation process or method are diagrammed in the state transition diagrams of
A state transition diagram is well understood by the skilled worker, and generally speaking, the arrows are transitions which occur upon the existence of a condition noted near its respective arrow. Thus, in
Threshold1 before the application of diversity. (Where the phrase “packet loss rate” is used herein, it is to be understood interchangeably with packet loss fraction or packet loss ratio which are similar concepts.) When Threshold1 is exceeded by packet loss rate F, a transition 101 occurs from the state (s11,d11) to state (s22,d22). When a gatekeeper request GK or Buffer occupancy full signal occurs, a transition 102 occurs from the state (s11,d11) to state (s21,d21). Gatekeeper and/or router boxes in the network may signal requests as just mentioned, and these are included in the state transition diagram for context and completeness of description.
Note that dotted ovals 111, 113, 115, etc. diagrammatically surround and thus indicate states that have the same sum of s and d components, and thus indicate essentially the same “overall transmission rate” (i.e., same network burden or load). From left to right in each of
In
Overall transmission rates: . . . (s11+d11)=(s12+d12)>(s21+d21)=(s22+d22)>(s31+d31)=(s32+d32) . . .
sij denotes the rate for Pn, and dij denotes the rate for Pn′.
In
In
Further in
Transitions like transition 101 from left to right in
Conversely, transitions like 103 and 105 from right to left in
Thresholds can be varying as well, such as depending on the source rate used. Thus, in
Further details of some more adaptation process embodiments are as follows:
1. Adapt both the source rate sij and the amount of diversity dij, in accordance with network fluctuations based on some QoS level measure (see examples in
2. Special case: Adapt both the source rate and the amount of diversity, in accordance with network fluctuations based on some QoS level measure and constrain their sum so that overall transmission rate is reduced or unchanged (see
3. As noted in 1 and 2, the process of increasing source rate is achieved through one or more stages. Two different steps which can either be used alone, or consecutively, or concurrently, are a) when increasing the overall transmission rate, maintain some diversity, and b) when reducing the amount of diversity, do not increase the overall transmission rate. Note that (a) and (b) can be realized using various combinations of source rate and diversity.
4. The approaches are applied to a) time diversity embodiments (media-specific redundancy, important information diversity, FEC, multiple description, multiple description data partitioning), b) path diversity embodiments, and c) combined time diversity and path diversity embodiments.
5. QoS level measure computations and Adaptation Logics.
5A. Delay-jitter handling via fixed-delay threshold embodiment declares a packet as lost, if the end to end delay of the packet is greater than a fixed threshold. Overall packet loss rate due to loss, delay, and delay-jitter (but before the application of diversity) is used as a QoS level measure. Thus, in
5B. Delay-jitter handling via adaptive packet playout embodiment performs delay-jitter handling using an improvement over S. B. Moon et al., “Packet audio playout delay adjustment: Performance bounds and algorithms,” ACM/Springer multimedia systems, January 1998. In this improvement, overall packet loss rate due to loss, delay, and delay-jitter (but before the application of diversity) is used as a QoS level measure.
Transition 101 of
(mode=SPIKE) OR (mode=NORMAL AND Overall packet-loss rate>Th1)
Transitions 103 and 105 respectively occur on a criterion (mode=NORMAL AND Overall packet-loss rate<Th2). Preferably but not necessarily, the overall transmission rate sij+dij is not increased on the transitions 101.
Here, the rate/diversity control 331 (or alternatively receiver 361′) detects whether network 351 is subject to spike-type delay increase and/or packet losses (SPIKE) even when the packet-loss rate has not yet exceeded the tolerable threshold Th1, or whether a more smoothly varying type of delay change/packet loss behavior (NORMAL) is occurring. This information is stored as a datum called “mode” for purposes of this embodiment 5B and used for adaptation. When SPIKE mode is occurring, the embodiment is relatively aggressive, being quick to initiate QoS-enhancing measures, and slow to end them.
One formula recognizes a SPIKE event when magnitude of delay difference of consecutive packets exceeds twice a variance measure+800 sampling intervals, compare the Moon et al. paper incorporated hereinabove at p. 21, Algorithm 2, line 2.
This 5B embodiment herein, however, not only recognizes a SPIKE event but also utilizes it for new purposes, processes and structures, to initiate a SPIKE mode to control a state machine of source rate and diversity amount. The SPIKE mode herein recognizes that packets arrive with an average delay based on the time of arrival minus the sender packet time stamp. Also, the packets have an average jitter magnitude, or measure of variance, in the varying delay values comparing packet to packet. The idea behind SPIKE mode herein recognizes an important control function for rate/diversity adaptation purposes when the magnitude of delay difference between consecutive packets exceeds some multiple of the measure of variance plus a constant. The multiple just-mentioned, reflects the idea that an onset of a significant delay difference in the incoming packets should be quite substantial compared to the usual amount of variation in delay in the packet stream. The constant reflects the idea that even if the measure of variance were equal to zero for a packet stream for a while, the onset of some delay difference would not be important if it were below the amount of the constant. It should be clear that various formulas and logic implementations can implement these ideas. One process embodiment determines when delay difference of consecutive packets |D(I,I−1)|>2J+800 to initiate the spike mode. Another process embodiment determines when delay difference
|D(I,I−1)|>mJ+c
to initiate the spike mode, where m is a numerical value of a multiplier selected in the range 1.5 to 4 for example, and the constant is equal to average measured delay based on timestamps for a last predetermined number (e.g. 25) of speech packets.
Another process embodiment uses a logic test to test whether the delay difference [|D(I,I−1)|>m1J] OR [|D(I,I−1)|>c1]. m1 is selected in the same range as numerical value m above. Constant c1 is suitably made substantially equal to constant c above.
Still another process embodiment uses any of the foregoing tests but with an average of delay difference magnitudes to smooth out the process somewhat, wherein
[|D(I+1,I)+|D(I,I−1)|]/2>mJ+c or alternatively a test
[[|D(I+1,I)|+|D(I,I−1)|]/2>mJ] OR [[|D(I+1,I)|+|D(I,I−1)|]/2>c].
Once the SPIKE mode has been initiated, then any one of various tests for returning to NORMAL mode is implemented. One embodiment repeatedly computes a measure of variance and waits until the measure of variance falls below a predetermined amount, whereupon the NORMAL mode is initiated. Still another approach utilizes the calculations of the Moon paper not only for playout delay, but also to derive controls for SPIKE mode and NORMAL mode in a manner that tracks the calculations of SPIKE and NORMAL conditions for playout delay as described in Moon et al.
5CA. A first (herein type 5C embodiment, subtype A) adaptation embodiment with parameters specified in RTCP uses both Fraction Lost and interarrival jitter field as QoS level measures having their respective thresholds.
Transitions 101 of
(Fraction Lost>Th1) OR (Interarrival Jitter J>Th2)
Transitions 103 and 105 respectively occur on a criterion (Fraction Lost<Th3) AND (Interarrival Jitter J<Th4)
Note in this 5CA embodiment that QoS enhancing measures are initiated on either an unacceptable level of Fraction Lost or of Jitter J. However, the QoS enhancing measures are relaxed on the occurrence of BOTH Fraction Lost and Jitter J becoming acceptable. Fraction Lost lower threshold Th3 is made less than or equal to Fraction Lost higher threshold Th1. Providing some gap between Th3 and Th1 may help prevent oscillations in QoS in some network environments. Similarly, Jitter lower threshold Th4 is made less than or equal to Jitter higher threshold Th2.
Preferably but not necessarily, the overall transmission rate sij+dij is not increased on the transitions 101.
5CB. As illustrated in
Transitions 101 of
(Fraction Lost>Th1) OR (Interarrival Jitter J increases for n consecutive RTCP reports)
Transitions 103 and 105 respectively occur on a criterion
(Fraction Lost<Th2) AND (Interarrival Jitter J is equal to or less than the Original Value)
Note in this 5CB embodiment, and in
Preferably but not necessarily, the overall transmission rate sij+dij is not increased on the transitions 101.
It should be apparent that numerous variations on this theme can be introduced in still further embodiment subtypes.
5D. Adaptation embodiment using TCP throughput estimate for both delay-jitter handling approaches compares a ratio of a corresponding TCP throughput estimate to current overall transmission rate with a threshold. One relatively uncomplicated TCP throughput estimate is given by: (constant×packetsize)/(round-trip delay×sqrt(average loss measured during the lifetime of the connection)). A suitable value of the constant is 1.22. See D. Sisalem et al., “The loss-delay based adjustment algorithm: A TCP-friendly adaptation scheme,” NOSSDAV, July 1998.
In
(TCP throughput estimate/current overall transmission rate<Th1)
As criterion for transitions 110, 103 and 105 this embodiment 5D uses (TCP throughput estimate/current overall transmission rate>Th2).
“sqrt” means the “square-root function of.” Still further, since thresholding is involved, the throughput estimate and current overall transmission rate can be squared, and compared with the square of threshold Th1 or Th2. This eliminates the square root calculation and speeds computation in some embodiments.
Preferably but not necessarily, the overall transmission rate sij+dij is not increased on the transitions 101.
By the use of diversity, the above embodiments handle short-term network fluctuations well, and cope with VoIP/VOP applications that involve multiple links of heterogeneous characteristics.
All the QoS level measure computations and adaptation logics are suitably used for rate/diversity adaptation (rate and diversity both, or either one alone, selected at different times, or selected on different transitions). All the QoS level measure computations and adaptation logics are suitably used for source rate adaptation alone in various embodiments without any diversity or diversity adaptation. All the QoS level measure computations and adaptation logics are suitably used for diversity adaptation alone in various embodiments without any source rate adaptation.
In
Preferably but not necessarily, the overall transmission rate sij+dij is not increased on the decreasing transitions 2211 and 2215 of
Overall transmission rates: . . . (s11+d11)=(s12+d12)>(s21+d21)=(s22+d22)>(s31+d31)=(s32+d32) . . .
Where (d11<d12), (d21<d22) and (d31<d32).
sij denotes the rate for Pn, and dij denotes the rate for Pn′.
Among various embodiments are embodiments for rate/diversity adaptation for Voice over IP and Voice over Packet. Described herein are systems, integrated circuits, and processes to adapt both rate and diversity, or each individually, in Voice over IP, Internet Audio, and Voice over Packet (VoIP/VOP) applications. Advantages include a robust solution for handling network impairments, while utilizing network resources efficiently.
As noted earlier, Voice over Packet (VOP) and Voice over Internet Protocol (VoIP) are sensitive to jitter to an extent qualitatively more important than for text data files for example. This sensitivity is also a problem for other types of real-time communication media such as frames of compressed video, but for brevity, VOP will be discussed as a placeholder for the other types of real-time communication as well.
The frame is the data unit for the speech coder. The packet can hold one frame or more than one frame. With constant number of frames per packet, packet loss rate is equal to frame loss rate.
ATM is a more sophisticated packet network wherein every packet in a stream takes the same path, so it represents a form of transmission that conceptually lies between circuit switching and packet switching. ATM, Frame Relay, and other forms of networking also can benefit by the improvements described herein.
RTP provides time stamps and packet sequence numbers. UDP (User Datagram Protocol) manages end-to-end transmission without any retransmission. UDP sits in the same layer as TCP. In one embodiment, RTP/UDP/IP is herein utilized for VOP instead of TCP/IP.
When multiple users congest the routers in the network, some packets become lost by actual loss or excessive delay.
In
The process executes a QoS determination step, which for example is a packet loss rate calculation over a predetermined window interval, given an expected rate of transmission. For example, if the connection protocol has identified a rate of transmission that is high, then a higher number of packets will be received during the same predetermined window interval at a given QoS than would be received the rate of transmission is lower. A simple packet loss rate calculation simply monitors the tags of the packets in a receive buffer, and counts up the number of packets that are present (a QoS measure) or those that are lost (Loss Rate which is just an inverse type of QoS measure). If a packet arrives in the buffer with a serial number and/or time stamp that indicates it is unusable, then it is dropped from the buffer and not counted. This is because VOP in some forms can only play or decode packets that have arrived in time to be meaningful to the user.
Another packet loss rate calculation that can be used with a short receive buffer keeps independently of the buffer a Service List of the packet tags and when they were received. The process simply counts from the Service List the number of packets which are within the window interval (e.g. last 5 seconds), or the number of missing packets depending on the approach.
Yet another process increments a counter when a new usable packet is received and decrements the counter when the time of arrival of a previously-counted usable but now-old packets has its time of arrival has become prior to the predetermined window interval from the present into the past. Other more sophisticated and arithmetically complex QoS measures are useful as well.
The skilled worker implements any suitable QoS determination. For example RTCP protocol has a reception block with a packet loss rate field wherein the protocol specification specifies how to compute a QoS measure at destination and transmit it back to the source. Thus, one type of embodiment suitably uses, supports or is compliant with RTCP protocol. Desirably, the source receives an “effective or overall packet loss rate or ratio” type of QoS measure which takes into account all lost packets, not only those actually lost in the network, but also those packets which came too late to be usable for the application, such as VOP, actually in use at the destination. Note further that the effective packet loss rate might be less when more sophisticated inventive VOP application software is implemented at the destination, even though the network congestion were no different.
Packets Lost=Packets Lost in Network+Packets Unusable at Destination.
EPLR=Effective Packet Loss Ratio=[Packets Lost in Network+Packets Unusable at Destination]/Packets Sent.
In
Both sides, source and destination, have speech encoder, rate/diversity control block, packet encapsulation, delay-jitter handling, lost packet compensation and speech decoder. Thus, it should be understood that for two way communication, there is suitably provided a transmit section 311′ (not shown) in the destination computer suitably (but not necessarily) identical to transmit section 311 described hereinabove. Also suitably provided is a receive section 361 (not shown) in the source computer suitably (but not necessarily) identical to receive section 361′ described hereinabove.
Primes in
Lost Packet block 381 (not shown) in the source also supplies via an RTCP packetizer 395 (not shown) second RTCP packet loss information descriptive of destination-to-source packet communication back via packet network 351 to the Control block 331′ in the destination.
In other more complex embodiments the path of communication from Lost Packet Compensation 381′ to Rate/Diversity control block 331 is suitably made independent of packet network 351, as by satellite, wireless, PSTN, etc.
Advantageously, control block 331 and compensation block 381 are each important improvements, singly and in combination with each other and in combination with the other blocks described. Also, feeding back STATE command information to a speech encoder improved to respond in its operations thereto, advantageously confers flexibility and control over QoS under different network 351 loading conditions.
Receive section 361′ has a Delay-jitter Handling block 371′ with a buffer in it, a process for reading the packet headers including their packet sequence numbers and time stamps, and a process of discarding or ignoring packets that arrive too late. Block 371′ is coupled to a Lost Packet Compensation block 381′ which utilizes any suitable means of reconstructing lost VOP data in lost packets such as by inserting zeroes or white noise, or by interpolation or by reconstructing from time-diversity, path diversity, or combined time/path diversity packet information.
In addition block 381′ calculates the QoS measure, such as packet loss ratio as described earlier hereinabove. Lost Packet block 381′ in the destination also supplies the RTCP packetizer 395′ the QoS measure which packetizer 395′ incorporates into the payload of return RTCP packets and sends them to control block 331.
Block 381′ couples commands and encoded speech data to speech decoder block 391′. The commands identify which of plural modes speech decoder block 391′ is to execute. For example, when only a single packet stream having a first type of encoding and transmission rate is being received, then the speech decoder is commanded to decode that first type of encoding and transmission rate. When a single packet stream having time-diverse packets in the stream is being received, then the speech decoder is commanded to decode by type of encoding, and to put the diverse packets information together to somewhat improve the quality of the output sound. When multiple packet streams having path-diverse packets are being received, then the speech decoder is commanded to decode by type of encoding, and to put the diverse packets information together to advantageously improve the quality of the output sound according to processes particularly emphasized herein. When multiple packet streams having not only time-diverse packets but also path-diverse packets are being received, then the speech decoder is commanded to decode by type of encoding, and to combine the packet information together to further advantageously improve the quality of the output sound.
Among various voice coders (vocoders) or speech coders contemplated for block 321 of
Turning again to
In connection with
RTP is suitably carried on top of UDP and IP. Each frame or set of frames of audio/voice/video/media has RTP header and a UDP packet contains the frame(s) and RTP header. The Payload type field in the RTP header identifies the type of coding that the encoder 321 uses.
RTCP is a control protocol for RTP. An RTCP “report packet” has a header as in
In RTCP report packet report block, the Fraction Lost field occupies eight (8) bits. Fraction Lost means the fraction of RTP data packets that were lost out of the packets sent by the described-sender since the last report packet was sent by the reporting sender. Fraction Lost is expressed as a binary fraction with the binary point at the left edge of the 8-bit field. Put another way, the integer occupying the 8-bit field is the Fraction Lost multiplied by 256. Put another way, Fraction Lost is the number of packets lost divided by the number of packets expected during the period since the last report packet. If the loss is negative due to duplicates, the fraction lost is set to zero. If all packets are lost in a reporting interval, no reception report is made. Note that in various alternative embodiments, the Fraction Lost calculation is either replaced by another QoS calculation, or suitably altered so that duplicates and diverse packets do not decrease the loss fraction.
In
The QoS level measure computation process (shorter line QoS in
Packets Lost=Packets Lost in Network+Packets Unusable at Destination.
EPLR=Effective Packet Loss Ratio=[Packets Lost in Network+Packets Unusable at Destination]/Packets Sent.
Number of packets lost in a time interval between reports is suitably calculated as the difference in the cumulative number of packets lost in the report packets. The ELSN (extended last sequence number) data in the report packets is used as follows. Calculate the difference in the ELSNs between two report packets to obtain the expected number of packets during the interval between the report packets.
The packet loss fraction PLR=Packets Lost in Network/#Packets Expected=
(Difference in Cumulative # of Packets Lost)/(Difference in ELSNs).
Note that the Effective Packet Loss Ratio includes not only Packets Lost in Network but also Packets Unusable at Destination in the numerator.
In
Interarrival Jitter J is a 32-bit mean deviation, smoothed absolute value, of the difference D(I, I−1) in packet spacing at the receiver compared to the sender for a pair of consecutive packets I and (I−1). Difference D is the absolute value of the difference in delays of at least two received packets. Delay d is the difference between a packet's RTP timestamp and the time of arrival in RTP time stamp units. In mathematical terms,
Delay d(I)=t(I)−s(I) (actual time received minus time stamp when sampled at source)
Delay Difference D(I,I−1)=d(I)−d(I−1)
J is suitably calculated in a calculation loop starting from an initialized value J of zero and successively calculating:
J=J+(|D(I,I−1)|−J)/N. N=16 is an example smoothing divisor constant used in RTCP.
Other approaches can calculate jitter as an average of absolute values of Delay Difference over a window. One procedure, among others suitable, is
J=J+[|D(I,I−1)|−|D(I-N,I-N-1)|]/N. N=16 is an example.
Still other approaches calculate jitter as the statistical variance, or otherwise suitably as the skilled worker elects for the purposes at hand.
So another type of embodiment computes jitter J in block 371′ to report back QoS. Jitter J reported back by RTCP is then compared to a threshold in rate/diversity control block 331 of
Still another type of embodiment computes in rate/diversity control block 331a joint function f (Loss Fraction, Jitter) and compares its value f with a threshold and then issues STATE controls based thereon according to a control loop similar to that described in
Other data in the RTCP report packet are suitably used in fashioning yet other embodiments.
Cumulative Number of Packets Lost is a 24 bit count of lost packets since beginning of reception.
Extended Highest Sequence Number Received is two data: First, the sequence number in the RTP header of the latest RTP data packet from sender 311 as received at receiver 361′. Second, a count of sequence number cycles.
Last Report Packet Time Stamp is the time of reception at receiver 361′ of the latest RTCP report packet received from sender 311.
Delay Since Last Report Packet is the time difference between reception of an RTCP report packet from the sender 311, and sending this RTCP report packet from the receiver 361′.
RTCP provides for Profile-Specific Extensions in the report packets. Therefore, various QoS functions as described herein can be computed at the receiver 361′ in block 371′ and put into the Profile-Specific Extensions area of the report packet in some embodiments. Otherwise, the QoS functions are suitably computed in block 331 of sender 311 from RTCP report packet information like Loss Fraction and Jitter coming back from the receiver. In still a further variation, embodiments use very short RTCP application-defined packets, called APP packets and receiver 361′ sends back these very short report packets instead of the longer RTCP report packets. The short APP packet suitably contains Packets Lost only, from which Loss Fraction is computed at the sender 311. Or the short packet suitably contains Cumulative Number of Packets Lost only. Or the short report packet from receiver contains Jitter and Loss Fraction only.
In this way, the introduction of block 381′ as a structural and process improvement into the system advantageously improves VoIP/VOP quality by establishing adaptive control of source rate sij and diversity dij. Block 381′ feeds QoS information such as packet-loss rate, delay statistics and any other information selected by the skilled worker as useful for this purpose, back to rate/diversity control 331. Rate/diversity control 331 thereupon responds to the feedback information according to any suitable process established in control 331 and described herein or hereafter devised to improve QoS when it becomes less satisfactory. Such process can operate according to a thresholding algorithm as described elsewherein, or respond in a more gradual manner either according to more closely spaced thresholds or according to a virtually continuous adjustment of source rate and diversity.
Further, the introduction of block 381′ as a structural and process improvement into the system advantageously improves VoIP/VOP quality by actually utilizing more of the packets sent to receiver 361′ via packet network 351. Block 381′ utilizes packets having diverse information and combines their information and controls speech decoder 391′ so as to form a speech output (or other audio or image or other media output) that more nearly replicates the speech input to speech encoder 321 originally or otherwise improves the quality.
Having a sender that has RTP protocol and a receiver that has RTCP to feed back a packet loss fraction to the sender are improved. The sender is improved by introducing rate/diversity control block 331 to add diversity and rate/diversity adaptation with state feedback to the speech encoder. Likewise improvements for lost packet compensation for the receiver are provided by block 381′.
Turning now to
Packet network 351 is a collection of interconnected routers or nodes, interconnected by links, and for the senders in a complex network, not all of the users are necessarily using the same links between the nodes. Thus, in
In
As can be seen from inspection of the
In
The
While in the rate diversity adaptation of
In
The amount of source rate adjustment and diversity adjustment responsively introduced by a given sender is subject to selection of any of various embodiment, the selection suitably made by the skilled worker bearing in mind principles of engineering economics, desirably short response-time, and other considerations.
For example, if the RTCP packet loss fraction datum rises above a tolerable Threshold1, one type of embodiment makes relatively smaller adjustments to source rate and diversity at the sender, and awaits reception of one or more additional RTCP packets to determine whether the packet loss fraction datum remains above the tolerable Threshold1, whereupon further adjustments to source rate and/or diversity are incrementally introduced until the packet loss fraction has been reduced acceptably.
In another type of embodiment, when the RTCP packet loss fraction datum rises above a tolerable Threshold1, such type of embodiment detects the difference between the packet loss fraction and Threshold 1 (or compares the fraction with Threshold1 and one or more additional even less tolerable higher Thresholds). Then block 331 in such embodiment makes adjustments to source rate and diversity at the sender, the adjustments being either incremental or more major depending on whether the packet loss fraction is near Threshold1 or in fact is much greater than Threshold1. Such embodiment does not await reception of one or more additional RTCP packets to determine whether the adjustment should be major rather than incremental. However, such embodiment does further update its adjustments and operations utilizing packet loss fraction data from further RTCP packets. The process in such embodiment is suitably tuned to produce adjustments that converge upon appropriate level of QoS in a desirably short response time. The process is tuned to prevent any major adjustments from leading to oscillation. Oscillation occurs when major decrease adjustments alternate with major increase adjustments, or divergent sequences of adjustments happen that do not contribute to satisfactory QoS.
Sender 311 decides whether to perform rate/diversity adaptation depending on the particular packet loss fraction (or other QoS measure) reported back from the particular receiver 361′ to which sender 311 communications are destined. Sender at node 301.q decides whether to perform rate/diversity adaptation depending on the particular packet loss fraction (or other QoS measure) reported back from the particular receiver at node 301.p to which the communications from the sender at node 301.q are destined.
The adjustments thus involve two or more of adjusting source rate sij, adjusting diversity rate dij (involving either or both of time diversity and path diversity rates), and adjusting the overall transmission rate sij+dij.
It is possible to have one receiver 361′ having no problem with packet loss and another receiver say at 301.p having a lot of packet loss. Advantageously, various embodiments avoid contributing in any way to an metastable or unstable network equilibrium wherein some nodes would interact collectively to “hog” network resources and others would be starved for network resources. Some schemes use redundancy wherein they repeat packets or information therein and thus increase the overall transmission rate. An important advantage of some embodiments is to use diversity adaptively—in other words, based on the network needs—and to avoid making congestion or the deleterious significance of a given bottleneck-link worse. Some embodiments have a first advantage of using the diversity as it is needed, and/or a second advantage of trying to reduce the congestion by both changing the source rate and the amount of diversity as well.
If some senders were encountering a lot of packet loss problems, then suppose that they start adding diversity and thus attempting to use more network resources and congesting the network some more. A further sender and receiver pair are now forced over their Threshold1 of tolerable packet loss fraction, due to the increase in network congestion due to the first-mentioned senders. Suppose the further and now newly over-Threshold1 sender and receiver also add diversity and attempt to use more network resources and congest the network still more, introducing QoS problems at further and further senders and receivers, in a snow-balling or chain reaction effect, wherein the packet network becomes even more greatly congested leading to network problems. Various embodiments avoid this problem by adding diversity to recover QoS adaptively, and either use the same overall transmission rate sij+dij, or decrease the overall transmission rate.
The response of sender 311 under control of block 331 might in some embodiments, as illustrated in
Rate and diversity usage of the network is optimized subject to the constraint that overall transmission rate be less than or equal to a given amount (which might change with network conditions). In optimization mathematics terminology, a QoS merit function is optimized subject to at least one constraint, namely that overall transmission rate be less than or equal to a given amount.
QoS is affected jointly by the amounts of source rate at which the speech encoder is transmitting and given by variable sij and diversity given by variable dij commanded by control block 331 of
The overall transmission rate is allocated between the source rate and the diversity dij.
Given a subsisting state of the network, QoS is a function of sij and dij. Mathematically, QoS is a function of two variables sij and dij. Graphically, QoS is a surface in a three-dimensional space having QoS as a vertical dimension and having two horizontal dimensions sij and dij. Thus,
QoS=f(sij,dij).
As between two users p and q, QoS(p,q,t) represents the QoS for communications from p to q at time t. QoS(q,p,t) represents the QoS for communications in the opposite sense from q to p at time t. QoS varies with number of users N, speech activity A, and over time depending on the configuration state of the network. Put another way,
QoS=QoS(p,q,sijp,dijp,A,N,t).
This 7-dimensional QoS function expresses the idea that QoS depends on which two users are involved, the source rate sij from sender p, the diversity rate dij from sender p, number of users N and the time t.
Let L be the packet loss after the application of diversity, meaning after any packet recovery or reconstruction that is implemented. Packet loss rate L is inversely related to QoS by a function g(L) so
QoS=g(L((p,q,sijp,dijp,A,N,t)).
Further, the following inequality expresses the challenge solved by embodiments herein to keep QoS above a threshold subject to a constraint on each sender p on overall transmission rate:
QoS(p,q,sijp,dijp,A,N,t)>threshold
sijp+dijp<=max overall transmission rate respective to each sender p.
or, put otherwise,
L(p,q,sijp,dijp,A,N,t)<threshold
sijp+dijp<=max overall transmission rate respective to each sender p.
The sender takes advantage of the diverse properties of the packet network so as to reduce the packet loss rate and increase the QoS to at least an acceptable amount.
When the threshold applies to all the users on the network, then the challenge of maintaining high QoS calls for the network-friendly advantages discussed elsewhere herein.
Some embodiments have just one QoS threshold wherein if the transmission quality becomes less acceptable than that threshold, an embodiment will make an adjustment in rate/diversity adaptation indicated by transition 101 of
Control block 331 takes original compressed speech packets P0, P1, P2, P3 and suitably adds diversity. If diversity robustness becomes needed, then it adds P0′ a form of the information in packet P0. So if the network loses the packet P0, or P0 comes too late and thus is not available while P0′ is on hand, then the lost packet compensation block 381′ uses bits P0′ to reconstruct the information in packet P0 to some extent or even fully.
In
Time diversity, in a second example has packet 621 with the information of packet P1 and trailing bits for compressed information dependent relative to two previous packets. Packet 621 is followed by packet 623 bearing information of packet P2 and an appended, or trailing, set of compressed bits having information dependent relative to packet P1 and the next previous packet before P1. Packet 623 is similarly followed by packet 625 bearing information of packet P3 and two trailing sets of compressed bits having information dependent relative to packets P2 and P1 respectively. Succeeding packets have information of the nth packet and two trailing sets of compressed bits having information dependent relative to the nth packet's two predecessor packets P(n−1) and P(n−2).
Further in
Another embodiment of time diversity has packets 641, 643, 645, etc. having information same as packets P1, P2, P3, etc. as well as respective appended information bits dependent as a joint function of the information in the two preceding packets, e.g. f(1,2).
In
In a further alternative, control block 331 maintains source rate s22 equal to s11 and diversity d22 is added as shown in a packet stream 1131. Here again overall transmission rate exceeds the original transmission rate of stream 1111.
In
In
In
For example, the TMS320C54x fixed-point, DSP family is fabricated with a combination of an advanced modified Harvard architecture which has one program memory bus and three data memory buses. This processor also provides a central arithmetic logic unit which has a high degree of parallelism and application-specific hardware logic, on-chip memory, additional on-chip peripherals. This DSP provides a specialized instruction set for operational flexibility and speed of the DSP.
Separate program and data spaces allow simultaneous access to program instructions and data. Two reads and one write operation can be performed in a single cycle. Instructions with parallel store and application-specific instructions are provided. Data can be transferred between data and program spaces. The parallelism supports a powerful set of arithmetic, logic and bit-manipulation operations that can all be performed in a single machine cycle. Control mechanisms manage interrupts, repeated operations and function calling. On-chip RAM and ROM memories are provided. Peripherals on-chip include serial port and HPI host port interface.
In
In
On a receive path in
Also, the delay and jitter handling portion of Packet Playout Control Unit 1581 includes software process steps to produce second lost packet fraction information representative of the incoming voice packets to integrated circuit 1511. This second lost packet fraction information is fed via a path 1585 to Packet Encapsulation Unit 1571 which packetizes the second lost packet fraction information into outgoing RTCP packets to update the destination across the network. The destination is suitably improved with an integrated circuit 1511′ (not shown) similar to or identical to integrated circuit 1511 of
From Packet Playout Control Unit 1581, depacketized compressed voice information being received is then supplied in a controlled manner to a speech decoder 1555 portion of speech coder 1541. Silence packets and voice packets, suitably dejittered and compensated by use of diversity packets as improved according to any of various process embodiments herein, then are decoded by speech decoder 1555 and thus played out. The speech thus played out, passes via Gain Control 1521 to PCM interface and from there to a DAC (digital to analog converter) not shown which can be provided either on-chip or off-chip as the skilled worker elects. The PCM output as converted by the DAC thus reconstitutes the voice in an advantageous manner more fully satisfactory and enjoyable to the user, by virtue of the various improvements provided and discussed herein. Further, a DTMF “touch-tone” generator 1591 and Tone Detector 1593 handle the dialing steps for placing a VOP/VoIP telephone call to confer a comprehensive application improved as discussed herein.
Operations of add diversity block 1561 of
Next in
In the decision step 1615, the value L is compared to determine if L exceeds a first threshold, designated Threshold1 or Th1, indicating too much packet loss at the destination. If too much packet loss, then operations proceed from step 1615 to a decision step 1617 to determine whether L also exceeds an even higher level A. If L is less than or equal to A, operations go next to a moderate update step 1621. If L exceeds A, operations go to an aggressive update step 1623.
In step 1621, a vector NEWSTATE is moderately updated in the manner shown in
If the source rate can be decreased no further, and the diversity can be increased no further, then step 1621 simply makes NEWSTATE the same as the current state. After step 1621 operations go to step 1651.
In step 1623, vector NEWSTATE is aggressively updated (for example to state (s32,d32)) in the manner shown in
Some embodiments have thresholds B, C, etc., such that NEWSTATE is moderately updated if (A≧F>Th1), NEWSTATE is aggressively updated if (B≧F>A>Th1), and NEWSTATE is even more aggressively updated if (F>B>A>Th1).
If in step 1615, L does not exceed Threshold1, operations proceed to a decision step 1627. Step 1627 determines if either the gatekeeper signal GK is on (GK=1) OR a buffer full flag is on (BFR=1). If NO, then operations go directly to step 1625, but if YES, operations go to an update step 1629.
In step 1629, vector NEWSTATE is updated (for example to state (s31,d31)) in the manner shown in
In decision step 1625, the packet loss fraction value L is compared with a second threshold Threshold2, or Th2. If value L is less than Th2, then QoS has improved or is at a high level already. In such case, operations pass to a step 1631 to update NEWSTATE to increase the source rate. Step 1631 inputs a new estimated steady state overall transmission rate S. As in steps 1621 and 1623, NEWSTATE is suitably updated according to any software method selected by the skilled worker, such as by looking up in a table, or executing a CASE statement, or implementing a software state machine, or otherwise. As indicated in
After step 1631 operations go to a step 1651. If in step 1625, value were greater than or equal to second threshold Th2, then operations go to decision step 1635.
In decision step 1635, the packet loss fraction value L is tested to see if it lies in the range from first threshold Th1 to second threshold Th2 inclusive. If yes, then operations go to a step 1641 wherein intermediate NEWSTATE is filled with the values in the vector STATE. Together with a later step 1651, this operation 1641 maintains the current control state. If the decision in step 1635 is NO (out of range), or when step 1641 is completed, then operations pass to step 1651.
In step 1651 the vector STATE is filled with the values of NEWSTATE. Next in an output step 1661, the values of STATE are output as control signals (sij, dij) to the encoder 321 of
The herein-incorporated patent U.S. Pat. No. 6,496,477 provides further disclosure about how path diversity packets are added. The emphasis of FIGS. 1,3, 15, 16, 17 and 18 are on the adaptive control features of some process, integrated circuit and system embodiments whereby source rate and diversity are either individually or jointly initiated, increased, decreased and terminated. The said adaptations are performed according to a process embodiment in response to QoS-related data obtained from the network or from a destination monitoring process. The adaptively-determined sij and dij, or controls generated in a relationship to sij and dij in the manner of a function thereof or substantially correlated to them, are then used to start, stop and adjust the operations of the diversity software and hardware disclosed in herein-incorporated U.S. Pat. No. 6,496,477 and the other disclosure herein.
In
Further in
In
In
In
A DSP interface manager software block 1821 is coupled to software blocks 1811, 1815, 1819 and 1823 and communicates with DSP 1511 and the software blocks described in connection therewith.
MCU 1781 runs system software 1823 including RTOS (real time operating system such as Microsoft Windows CE or Symbian EPOC, as well as DSP BIOS™ RTOS from Texas Instruments Inc.) System software 1823 includes WDT driver software, flash memory manager, BSP software, development and self-test (IPQST) software, and software installation code.
DSP 1511 has software in
In other embodiments, as shown in
Some of the cell-phones 1921, 1923, 1925, and 1937 have a shaded rectangle, indicating for purposes herein improvements for adaptive rate/diversity as disclosed herein in their packet voice communications mode, for example as shown in any one or more of
Some of the cell phone base stations 1911, 1913 have a shaded rectangle, indicating for purposes herein improvements for adaptive rate/diversity as disclosed herein in their packet voice communications mode, for example as shown in any one or more of
Personal computer (PC) telephony units 1951 and 1953 have respective microphones and speakers, and these units 1951 and 1953 have modems of any suitable type, such as voice-band V.90, DSL (digital subscriber line), cable modem, wireless modem, among other choices. Personal computer (PC) telephony units 1951 and 1953 are respectively coupled to the network 351 via gateways 1961 and 1963 respectively. The gateways are suitably located in a private branch exchange or in a telephone central office, or in the office of an ISP (Internet Service Provider) or in the office of a private commercial network, for example. IP-packetization occurs at the PC telephony units 1951 and 1953. The adaptation is end-to-end such as when phone at source has the rate/diversity control block and the phone at destination has a block to send QoS data back as well as to couple diverse packet information to the decoder for improved QoS.
For placing phone calls over the Internet, user voice goes in through microphone, then is processed in the computer by the main microprocessor, microcontroller, and DSP for vocoding and rate/diversity adaptation as in
Adaptive rate/diversity improvements in an integrated circuit, software and system are suitably provided in an Internet mobile terminal such as an Internet appliance or mobile phone, cell phone or cordless phone with Internet or other packet network capability.
Cell phone base stations 1915, 1917 and 1911 are respectively coupled to IP packet network 351 via PSTN blocks 1971, 1973 and 1975 respectively. Each of the PSTN blocks 1971 and 1973 has a gateway therein to connect the call to the packet network 351. The gateways in PSTN blocks 1971 and 1973 suitably have adaptive rate/diversity embodiments included therein. Thus, rate/diversity adaptation modules suitably are sited in the gateways and base stations of the system of
A gateway GW 1981 couples network 351 to PSTN 1983 to which telephones (not shown) are coupled via a PBX 1985. Also, one or more individual telephones 1987 are directly connected to PSTN 1981. Further in
Integrated circuits into which the adaptive rate/diversity improvements are suitably manufactured include DSP (digital signal processor) from Texas Instruments and other companies offering DSP integrated circuits. Other integrated circuits suitable for the adaptive rate/diversity improvements include host microprocessor such as Intel's Pentium.RTM., Pentium II.RTM., Pentium III.RTM., Celeron.RTM., Xeon.RTM. and IA-64 microprocessors, AMD K6 and K7 microprocessors, National MediaGX and other microprocessors, and microcontrollers such as ARM and StrongARM series, MIPS series, Intel i960, Motorola Mcore and PowerPC integrated circuits, among many others. Still other integrated circuits which are contemplated for adaptive rate/diversity improvements include nonvolatile memories such as ROM, EPROM, EEPROM, Flash memory, EAROM, and FeRAM (Ferroelectric random access memory). Volatile memories such as DRAM, synchronous DRAM (SDRAM), R-DRAM (Rambus DRAM), DDR-DRAM, and other variants suitably also have a logic section or non-volatile section incorporating the adaptive rate/diversity improvements built into them as taught herein. In yet other embodiments, the adaptive rate/diversity improvements are loaded onto or manufactured into rigid disk drives, hard disk drives, and also various media such as floppy insertable disks, CD-ROM optical storage media, and/or chips in the read circuitry or other circuitry of drives for such storage media. Also, chipsets associated with processors suitably are in improvement embodiments made to have adaptive rate/diversity improvements manufactured into them, such as the Intel “440xx” series of chipsets, sometimes known as North Bridge and South Bridge chips, and the chipsets of other chipset manufacturers. (Chipsets of this type are also suitably improved with digital signal processors, as taught in any one or more of the patents U.S. Pat. No. 5,987,590 and U.S. Pat. No. 6,179,489 which are hereby incorporated by reference. The DSP suitably runs the adaptive rate/diversity improvements. In other versions, the adaptive software runs on the host microprocessor such as Pentium series or IA-64 series, or partitioned with part of the software on a DSP coupled to the host microprocessor(s) in the computer system.
Also, the rate/diversity adaptation can be provided at the telephone central office gateway. Also, rate/diversity adaptation can be put in a router in a packet network to improve it there.
Even more advantageously, when the adaptation is end-to-end and the units at both ends have at least the adaptation software, the mobile phone or desktop or notebook PC telephony unit adapts for advantageously satisfactory QoS.
As discussed further, an improved cell-phone base station (and also a gateway improved similarly) runs multiple packet voice modules with an inventive embodiment for each mobile telephone using the base station at a given time.
Some improvement embodiments are intended for gateways, wherein a improved gateway embodiment runs multiple packet voice modules with an inventive process, chip and system embodiment working in the gateway itself to adaptively control source rate and diversity rate for advantageous QoS for each telephone using the gateway at a given time.
In other embodiments, a base station itself is not only improved to be multimodal, supporting both the mobile Internet Protocol phones (IP-phones) and conventional mobile phones. But also, the improved base station embodiment runs multiple packet voice modules with an inventive process, chip and system embodiment working in the base station itself to adaptively control source rate and diversity rate for advantageous QoS for each cell-telephone communicating speech in non-packet form to the base station at a given time. Then the base station itself and not necessarily the cell-phone codes or recodes the speech and packetizes it with an inventive process, chip and system embodiment working to adaptively control source rate and diversity rate for advantageous QoS over a packet network to which the base-station is in turn connected.
Numerous combination embodiments and paths of advantageous operation are conveniently identified in
In
When operations are in state (s11,d11), e.g. (16.0, 0.0) and either a gatekeeper request GK=1 or Buffer Occupancy BFR=1 occurs, then adaptive source rate measures are employed without diversity measures, in this example. In such case, a transition 2215 goes from (s11,d11) to state (s31,d31), e.g., (8.0,0.0) kbps of source rate and no diversity rate. After transition 2215, operations remain at state (s31,d31) unless and until both the gatekeeper request is turned off and the Buffer Occupancy condition is not present, i.e. GK=0 AND BFR=0. At that point, state transition 2231 takes the system from state (s31,d31) to state (s21,d21) e.g., (11.2,0.0). By transition 2231, source rate is increased, and diversity remains off. Operations at state (s21,d21) poll the gatekeeper and buffer for updated status information. Then operations remain at state (s21,d21) unless the GK and BFR remain off, or in case GK or BFR go on again and later become both off, i.e. GK=0 AND BFR=0. Thereupon a state transition 2233 transfers the system to state (s11,d11) e.g., (16.0,0.0). By transition 2233, source rate is increased, and diversity rate remains disabled or at zero, and their sum (overall transmission rate) is increased. Operations remain at state (s11,d11) unless criterion F causes aggressive transition 2211 of
The processes and systems of
Note that the state transition diagrams of
The adaptive rate/diversity improvements are suitably implemented in conferencing, broadcast, unicast, and multicast devices and processes since UDP universal datagram protocol, RTP real-time transport protocol and RTCP and other protocols now available or yet to be devised are useful for supporting these services.
Broadcast with path diversity is described in connection with incorporated patent U.S. Pat. No. 6,496,477
Multicast with path diversity is described in connection with incorporated patent U.S. Pat. No. 6,496,477
Before proceeding further, note an example process context in
The operations of steps 2441 and 2431 are next described in considerable detail. Note that there are many alternative ways of doing each of them, and an outline format is used to facilitate the detailed description.
Accordingly, several embodiments of adaptive rate/diversity control of improved multicasting are contemplated. As noted hereinabove, rate/diversity adaptation of a given media stream at the source 1111 of U.S. Pat. No. 6,496,477 FIG. 12 affects plural destinations. QoS reports come back to source 1111 from the various destinations for each portion in a series of portions comprising the transmission from source 1111. The QoS reports from the various destinations for a given portion of the transmission are combined into one or more herein-defined “Multicast QoS” evaluation numbers in
A. In a first method, the QoS reports from the various destinations for a given portion of the transmission are combined into a single herein-defined “Multicast QoS” evaluation number in step 2441 to drive the rate/diversity adaptation processes. In other words, Multicast QoS (MQoS) is defined for each corresponding transmission portion such as activity in a 5-second interval described by an RTCP report packet. The step 2431 of
A1. The MQoS in step 2441 depends on what happens to fewer than all of the destinations
A1a. The MQoS depends on what happens to a majority of the destinations. Example:
For the 37 th RTCP packet, QoS values came back from 150 out of 155 destinations. The A1a embodiment is programmed to find the value of a statistic based on the QoS values from the best-QoS reporting X % of the 155 destinations, e.g., (say 90% of them), the best 140 (=155×0.90) out of the 150 QoS values.
A1ai. MQoS=average of the best reporting X %
A1aii. MQoS=minimum of the best reporting X %
A1aiii. MQoS=median of the best reporting X %
A1aiv. MQoS=average of the worst reporting Y %
A1av. MQoS=minimum of the worst reporting Y %
A1avi. MQoS=median of the worst reporting Y %
A1avii. MQoS=maximum of the worst reporting Y %.
A1b. The MQoS depends on what happens to selected ones of the destinations. Example: For the 23rd RTCP packet, QoS values came back from 205 out of 324 destinations. The A1b embodiment is programmed to find the value of a statistic based on the QoS values disregarding the best-QoS reporting X % of the 324 destinations and disregarding the worst-QoS reporting Y % of the 324 destinations. MQoS=average of 20 percentile to 90 percentile loss fraction RTCP reports.
A1c. The MQoS depends on randomly selected ones of the destinations. Example: For the 147th RTCP packet, QoS values came back from 2000 out of 3000 destinations. The A1b embodiment is programmed to find the value of a statistic based on a random sample of N=100 of the 2000 QoS values. Then the processes apply a statistical computation according to any of the following alternatives:
A1ci. MQoS=average of the N selected QoS values
A1cii. MQoS=minimum of the N selected QoS values
A1ciii. MQoS=median of the N selected QoS values
A1civ. MQoS=average of the selected QoS values, but leaving out their top X % and bottom Y %.
A2. A rate/diversity adaptation decision step depends on a MQoS statistic based on all the destinations' reports of QoS received back for the given transmission portion.
A2a. The statistic is the median QoS. Example: For the 7 th RTCP packet, QoS values came back from 150 destinations out of 180 destinations. The loss fraction values varied from 0.2% to 12%, mostly around 3%. 2.8% loss fraction was the median value. MQoS=2.8%.
A2b. The statistic is the QoS of the destination at the nth percentile of QoS. Example: For the 17th RTCP packet, QoS values came back from 138 destinations out of 180 destinations. The embodiment is programmed to find the 30th percentile as indicated by listing the reports in loss fraction order. Example: The loss fraction values varied from 0.2% to 12%, mostly around 3%. The 30th percentile value was 6.2%. MQoS=6.2%.
A2c. The statistic is the average QoS. Example: For the 24th RTCP packet, QoS values came back from 155 destinations. The embodiment is programmed to find the arithmetic mean or average QoS. Example: The loss fraction values varied from 0.2% to 12%, mostly around 3%. The arithmetic mean was 3.3%. MQoS=3.3%.
A2d. The statistic is the minimum QoS of any destination. Example: For the 33rd RTCP packet, QoS values came back from 125 destinations. The loss fraction values varied from 0.2% to 12.2%, mostly around 3%. The minimum QoS was 12.2% loss fraction. MQoS=12.2%.
In a second process type, the QoS reports from the various destinations for two or more portions of the transmission are combined into a single herein-defined “Multicast QoS” evaluation number to drive the rate/diversity adaptation processes. Two or more RTCP packets from the same destination are used to generate each report datum by averaging, median, minimum or other statistic, and step 2431 becomes more detailed. If only one RTCP packet from a given destination comes back when a most destinations are reporting back three for the given process, then the report datum is the value of that RTCP packet. The report data thus derived destination by destination are used according to any of the A-numbered processes to generate MQoS by step 2441 of
Example Embodiment B2a: The statistic is the median QoS. Example: For the 7th through 12th RTCP packets, QoS values came back from 150 destinations. Each set of six RTCP packet Loss Fraction values for a given destination was averaged to produce a report datum for that destination. The report data varied from 0.5% to 7%, mostly around 3%. 2.9% loss fraction was the median value. MQoS=2.9%.
A tedious description of other B-type embodiments is suitably generated by following the directions of the previous paragraph, which is believed to amply disclose the subject matter of numerous B-type embodiments.
Other QoS level measures and adaptation logics are suitably combined with the teachings and figures shown herein.
If in step 2511 the diversity flag is already on, then operations branch to a step 2551 to vary the source rate and diversity aspects of the coder. Then in step 2561, the packet header is updated in the dependency information and diversity fields to correspond to the changes made by step 2551, whereupon RETURN 2571 is reached.
In
If in step 2621, the overall transmission rate is below value S, then operations branch to a step 2661 to change both the source rate and diversity (and suitably the path diversity method of the Packet Transmission Table of U.S. Pat. No. 6,496,477 and other diversity methods) without closing down the diversity feature. Then operations go to step 2651 to update packet header as described above.
If in step 2611, the diversity flag is off, then operations branch to a step 2671 to increase the source rate only, whereupon RETURN 2681 is ultimately reached.
In
In a complementary way, the network advantageously controls overall transmission rate, so that transitions between ovals in
In
If at any 11.2 kbps state estimation EST=8, then a transition goes from the originating 11.2 kbps state to state (4.0,1.7,2.3). If GK=0 AND BFR=0 is signaled by the network, then a transition goes from any originating 8 kbps state back to state (5.7,2.3,3.2) provided that a ratio R is also less than or equal to a threshold Th3. For example, R is the ratio of estimated steady state overall transmission rate divided by current overall transmission rate. Th3 suitably lies in a range of 1.5 to 4.0, and a value of 3.0 is suitable. Estimated steady state overall transmission rate S is that rate which the network signals is now available or which test algorithms at sender or receiver indicate is now available. Rate S is suitably computed as in the discussion of
If at any 11.2 kbps state, GK=0 AND BFR=0 is signaled by the network, then a transition goes from the originating 11.2 kbps state to state (8,3.2,4.8). If at any 8 kbps state, network conditions indicate greatly lessened congestion, then an aggressive recovery to state (8,3.2,4.8) is desirable. In
Within a given oval, DSP software determines from QoS reports whether to make transitions to add diversity, relax diversity, or release diversity. Starting at state (16,0), a determination that A≧F≧Th1 adds diversity and takes operations to a state (11.2,4.8). However, starting at state (16,0), if F>A>Th1, then operations adds two stages of diversity and goes to a state (8.0, 3.2,4.8). Starting at state (11.2,4.8), a determination that F>Th1 adds further diversity and takes operations to state (8.0,3.2,4.8). If QoS becomes ameliorated, such that F<Th2, then operations are transitioned from state (8.0, 3.2,4.8) to state (11.2,4.8) and/or from state (11.2,4.8) to state (16,0) as shown.
Starting at state (11.2,0), a determination that A≧F>Th1 adds diversity and takes operations to a state (8.0,3.2). However, if F>A>Th1, then starting at (11.2,0) operations become more aggressive and add two stages of diversity and go to a state (5.7,2.3,3.2). If F>Th1 continues (unacceptable QoS) at state (8.0,3.2), then operations add diversity and go from state (8.0,3.2) to the state (5.7,2.3,3.2). If QoS becomes ameliorated, such that F<Th2, then operations are transitioned from state (5.7, 2.3,3.2) to state (8.0,3.2) and/or from state (8.0,3.2) to state (11.2,0) as shown.
Starting at state (8.0,0), a determination that A≧F≧Th1 adds diversity and takes operations to a state (5.7,2.3). However, if F>A>Th1, then starting at (8.0,0), operations become more aggressive and add two stages of diversity and go to a state (4.0,1.7,2.3). If F>Th1 continues (unacceptable QoS) at state (5.7,2.3), then operations add diversity and go from state (5.7,2.3) to a state (4.0,1.7,2.3). If QoS becomes ameliorated, such that F<Th2, then operations are transitioned from state (4.0,1.7,2.3) to state (5.7,2.3) and/or from state (5.7,2.3) to state (8.0,0) as shown.
With the use of ratio R in the transition criteria, overall transmission rate changes from low rate ovals are advantageously arranged to be larger than rate changes on return transitions between higher rate ovals. Thus, successively smaller increases in rate are achieved with finer increases as higher rates (and attendant network burden) are approached.
The histogram information is here recognized as quite useful for rate/diversity adaptation purposes. Even though the packet loss rate might be the same in two different cases, the aggressiveness of adaptation measures is suitably made more aggressive if the histogram is more populated with higher numbers of packets lost consecutively.
In
If z2 occurs, then operations instead go from state (s11,d11) to reduce source rate and introduce two amounts of diversity d42,e42 at state (s42,d42,e42). This is the aggressive adaptation.
Criterion z1 is suitably established as (A≧F>Th1) AND frequency of two-consecutive-losses-or more is Th2 or less.
Criterion z2 is suitably established as (F>A>Th1) OR frequency of two-consecutive-losses-or more exceeds Th2.
If and when QoS becomes ameliorated, such that a criterion z3 is met, then operations are transitioned from state (s42,d42,e42) to state (s22,d22), and/or from state (s22,d22) to state (s12,d12), and/or from state (s12,d12) to state (s11,d11) as shown.
Criterion z3 is suitably established as (F<Th3) AND frequency of two-consecutive-losses-or more is less than a threshold Th4.
Th1 is suitably made 3%, Th2 is suitably 2%, Th3 is suitably 0.5% and Th4 is suitably 0.25%. The skilled worker suitably tunes the thresholds. Also, an automated tuning process suitably varies the thresholds over illustrative ranges 1-5% for Th1 and Th2, and over 0% to 2% for Th3 and Th4 for most satisfactory adaptation operation.
Note among other advantageous features of the process of
Bracketed sets of packets illustrate the meaning of each state. State 3211 corresponds to transmission of packets in a series of packets each with a header H and a payload comprising one frame of compressed data, sent at a certain number of packets per second and a certain number of frames per second.
A second state 3221 involves transmission of packets in a series of packets each with a header H and a payload comprising two frames of compressed data sent suitably (but not necessarily) at the same number of frames per second as in state 3211, but at a different and fewer number of packets per second. Notice that a brace indicates 3 payload frames corresponding to comparable information distributed differently among packets depending on which state is used.
A third state 3231 involves transmission of packets in a series of packets each with a header H and a payload comprising three frames of compressed data sent suitably (but not necessarily) at the same number of frames per second as in state 3211, but at a different and still fewer number of packets per second. Notice that 3 payload frames are included in the same one packet with its one header when state 3231 is used.
Return transitions occur when criterion F<Th2. One return transition takes operations from state 3231 to state 3221. Another return transition takes operations from state 3221 to state 3211.
Note that the criteria for transition are suitably selected according to any of the various embodiments elsewhere described herein.
The variable frames-per-packet embodiments are suitably augmented with time or path or combined time/path diversity as shown in
Return transitions occur when criterion F<Th2. One return transition takes operations from state 3335 to state 3325. Another return transition takes operations from state 3325 to a state 3315 (1 frm/pkt, 8 kbps source rate, 8 kbps diversity rate). Another return transition takes operations from state 3315 to state 3211.
Flow diagrams of some processes for control of frames/packet are the same as
Again, other criteria for the transitions as described elsewhere herein are suitably employed. Each of the types of time diversity, path diversity, and time/path diversity as described herein and in the incorporated U.S. Pat. No. 6,496,477 are contemplated for use in various embodiments.
Note that changing number of frames per packet, it may be advisable in some embodiments to make the frm/pkt transition only during a silence period following a talkspurt featuring unacceptable QoS. Other embodiments suitably make frm/pkt transition during a talkspurt without restriction.
Another embodiment performs a hybrid frame/packet adaptation: frame/packet increase occurs during both silence periods and active speech, but frame/packet decrease occurs during silence periods only. Steps 2515, 2551, 2661 and 2671 are correspondingly improved by preceding them with tests for presence of a talkspurt flag or a silence flag, so that the transitions occur according to the just mentioned logic embodiments that depend on silence only, or talkspurt, or during either silence or talkspurt. If the required test is not met, the respective step 2515, 2551, 2661 or 2671 is bypassed, and if the test is met the respective said step is performed.
Also, note a possible effect on some diversity methods when changing number of frames per packet. Suppose, for example, that a time diversity P(n)P(n−1)′ in one packet is changed to P(n)P(n−1)′P(n+1)P(n)′ by changing to more frames per packet. If the longer packet is lost, both P(n)′ and P(n) are lost, meaning that all of the nth information is lost. Accordingly, some embodiments suitably change to a diversity method that is resistant to packet loss concurrently with (or at least close in time to) a transition from one number of frames per packet to a higher number of frames per packet.
Gateways, wireless base stations, private branch exchanges, networked appliances and other applications are suitably enabled by adaptive rate/diversity controls, chips, chipsets, printed circuit cards, and subsystems disclosed herein. Recoder and/or transcoding processes recodes or transcode the information and produces an output compressed and coded according to a different form than was received by a given device. It is contemplated that devices, processes and systems are suitably cascaded and integrated for various telecommunication and networking purposes. Where many channels are processed simultaneously, the systems are suitably replicated or multiplexed to the extent desired, so that software and hardware are effectively, efficiently and economically employed. Where blocks are shown herein, they are suitably implemented in hardware, firmware or software in any combination. The embodiments described are merely illustrative, while the scope of the inventive subject matter is defined by the claims and equivalents thereof.
This application is a divisional of prior application Ser. No. 10/815,044, filed Mar. 30, 2004, now pending; which was a divisional of prior application Ser. No. 09/460,065, filed Dec. 14, 1999, now U.S. Pat. No. 6,744,757, granted Jun. 1, 2004. The following coassigned patents are hereby incorporated herein by reference: U.S. Pat. No. 6,496,477, issued Dec. 17, 2002 U.S. Pat. No. 6,421,527, issued Jul. 16, 2002 U.S. Pat. No. 5,987,590, issued Nov. 16, 1999 U.S. Pat. No. 6,179,489, issued Jan. 30, 2001 The following patents relate to this application: U.S. Pat. No. 6,757,256, issued Jun. 29, 2004; U.S. Pat. No. 6,804,244, issued Oct. 12, 2004;U.S. Pat. No. 6,678,267, issued Jan. 13, 2004; U.S. Pat. No. 6,574,213, issued Jun. 3, 2003; U.S. Pat. No. 6,765,904, issued Jul. 20, 2004; U.S. Pat. No. 6,801,499, issued Oct. 5, 2004; and U.S. Pat. No. 6,801,532, issued Oct. 5, 2004.
| Number | Date | Country | |
|---|---|---|---|
| Parent | 10815044 | Mar 2004 | US |
| Child | 12494998 | US | |
| Parent | 09460065 | Dec 1999 | US |
| Child | 10815044 | US |
