Method And Apparatus Of Fully Distributed Packet Scheduling For A Wireless Network

Abstract

Disclosed is a method and apparatus for fully-distributed packet scheduling in a wireless network. The decoding algorithm with low-density parity-check code is applied in a transmission wireless network to achieve the fully-distributed packet scheduling. In the packet scheduling, only one wireless network node is needed to exchange information and communicate with its neighboring network nodes. Therefore, it is not necessary to estimate the signal to noise ratio, while being eye to eye among the neighboring network nodes. If the network load exceeds the network capacity, the present invention automatically eliminates the most difficult user to reduce the overall network load and diverts the resources to the surviving users.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1A shows a schematic view of a conventional single cell multihop wireless network.

FIG. 1B shows a schematic view of a conventional multi-cell multihop wireless network.

FIG. 2 shows a flowchart of the full-distributed packet scheduling method of the present invention in a relay network.

FIG. 3A shows a factor graph of FIG. 1A.

FIG. 3B shows a factor graph of FIG. 1B.

FIG. 4A shows the factor graph of FIG. 3A with information on the packet queue length of each network link, using downlink for communication.

FIG. 4B shows the factor graph of FIG. 3A with information on the packet queue length of each network link, using uplink for communication.

FIG. 5 shows a schematic view of an apparatus realizing FIG. 3.

FIG. 6 shows the comparison between the present and the RR and IP in terms of average transmission rate in each packet slot with uplinks for communication in a single cell multihop wireless network.

FIG. 7 shows the average transmission rates of mobile stations in each packet slot with uplinks for communication in a multicell multihop wireless network.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

In FIG. 1A, it can be observed that the packet scheduling problem can be transformed into a decoding problem for low-density parity-check (LDPC) code as long as the packet scheduling problem can be described by a factor graph model, and the constrain rules are defined for the variable vj on each network link. A standard process commonly known as sum-product algorithm can solve the decoding problem for LDPC code. To fully explore the network resources, the present invention designs a weighting scheme so that the user with more packets has a higher priority for transmission.

FIG. 2 shows a flowchart of the fully-distributed packet scheduling method of the present invention. First, step 201 is to construct a factor graph to model a multi-hop wireless network. The factor graph includes a plurality of agent nodes, a plurality of variable nodes and a plurality of edges. The factor graph is a mutually-interactive graph. Step 202 is to transform the packet scheduling problem into a decoding problem for LDPC code based on the factor graph, and use a standard process to solve the decoding problem. In other words, the sum-product algorithm is used as a tool for an access point (AP) to communicate with the neighboring APs. Step 203 is to apply a weighting scheme on the packet schedule according to the network condition at that time.

The construction of factor graph in step 201 can be implemented with the following three steps.

(a) An agent node is used to represent each network node. Each agent node is marked with the corresponding constrain function f_i, which defines an interference-avoiding local constrain rule.

(b) A variable node is used to represent a network link. Each variable node is marked with a variable v_j.

(c) Each variable node is linked to two agent nodes, where the network nodes corresponding to the two agent nodes can communicate through the network link corresponding to the variable node.

To improve the utilization of the network resource, the present invention uses the link communication to repeatedly exchange the soft-information of the probability mass function (pmf) of each variable node between the neighboring network nodes and variable nodes.

Taking FIG. 1A as an example, FIG. 3A shows a factor graph of a single cell network constructed by using the above three steps. Taking FIG. 1B as an example, FIG. 3B shows a factor graph of a multi-cell network constructed by using the above three steps.

To meet the basic multi-hop wireless network and the interference-avoiding local constrain rules, each network node must obey the following rules during each packet slot.

(a) A network node can only transmit to one network node during transmission.

(b) A relay cannot transmit and receive packets simultaneously.

(c) A network node cannot receive packets from multiple sources simultaneously.

In other words, when executing the interference-avoiding local constrain rule and sum-product convergence, each agent node in the example in FIG. 3A must obey the following constrain rules:

f ₁ : v ₁ +v ₂ +v ₃ +v ₄≦1; and

f ₂ : v ₁ +v ₅ +v ₆ +v ₈≦1; f₃: v₃+v₇+v₉+v₁₀≦1; and

f ₄ : v ₅≦1; f₅: v₆≦v₇; and

f ₆ : v ₂ +v ₈ +v ₉≦1; f₇: v₄+v₁₀≦1.

That is, the variables surrounding each agent node form a valid local 10 transmission pattern. In the example in FIG. 3A, the valid local transmission pattern is: for BS, {v₁, v₂, v₃, v₄}={(0,0,0,0), (1,0,0,0), (0,1,0,0), (0,0,1,0), (0,0,0,1)}; and for M3, {v₂, v₈, v₉}={(0,0,0), (1,0,0), (0,1,0), (0,0,1)}, and so on. The collection of all the valid local transmission patterns forms a valid global collision-free schedule.

It is worth noticing that the interference-avoiding local constrain rules can be applied to both single cell and multi-cell multihop wireless network.

The following describes how to compute and transport the soft-information that can improve the network resource utilization. The soft-information shows the probability that each network link will be utilized in each packet slot.

First, the probability P_b(v_j) of each variable node v_j (v_j=b, b is 0 or 1) is initialized. Then, the soft-information SI_st(x, y, b) of each agent node (marked by the corresponding constrain function f_i) linking to each variable node v_j, where P₁(v_j) is uniformly distributed between (0,1), and P₀(v_j)+P₁(v_j)=1. SI_st(x, y, b) represents the soft-information transported from node x to node y, and indicates the probability when the corresponding variable node v_j is b. The subscript s is the packet slot index, t is the iteration index. The initialized probability P_b(v_j) is the probability randomly assigned to each variable node v_j.

It is worth noticing that the initialized P_b(v_j) can be added through the network link corresponding to the variable node v_j to one of the two agent nodes linking the variable node.

Then, according to the standard sum-product algorithm, the soft-information SI_st(x, y, b) transported from variable node v_j to agent node (mark by the corresponding constrain function f_i) can be computed. Taking variable node v₁ and agent node f₂ as an example, the soft-information is computed as SI_st(v₁, f₂, b)=c_1,2 P_b (v₁)·SI_(s-1)t (f₁, v₁, b), where c_{1, 2} is a normalizing factor to make SI_st(v₁, f₂, 0)+SI_st(v₁, f₂, 1)=1.

Furthermore, in addition to weighting each valid local transmission pattern, the same sum-product algorithm is used to compute the soft-information transported from agent node (mark by the corresponding constrain function f_i) to variable node v_j. The agent node (mark by the corresponding constrain function f_i) collects all the incoming soft-information from its neighboring variable nodes {V_h}, and computes the soft-information SI_st(f_i, v_j, b) to be transported to variable node v_j. Taking BS as an example, the soft-information is computed as SI_st(f₁, v₁, 1)=d_1,1 {ω_t(f₁, 1000))·SI_st(v₂, f₁, 0) SI_st(v₃, f₁, 0) SI_st(v₄, f₁, 0)}, where d_1,1 is a normalizing factor.

The total soft-information SI_st(v_j, b) of variable node v_j can be computed as the product of all the soft-information SI_st(f_i, v_j, b), where f_i belongs to the set of all the agent nodes linking to variable node v_j.

It is worth noticing that the above soft-information computation of variable node v_j can be applied to both single cell and multi-cell multihop wireless network. After the second iteration, if SI_st(v_j, 1)≧SI_st(v_j, 0), variable node v_j is determined to be active; that is v_j=1; otherwise, v_j=0. When all the variable nodes v_j follow the interference-avoiding local constrain rule, the standard sum-product algorithm terminates and outputs a valid global schedule. Otherwise, the agent nodes not following the interference-avoiding local constrain rule must repeat the above sum-product computation.

A multihop wireless network may prefer a certain interference-avoiding or collision-free packet scheduling method to others because the former can maximize the reuse of the network resources. For example, in a single cell multihop wireless network, the service provider can prefer the global schedule {v_j}={0,1,0,0,1,0,0,0,0,1}, instead of the global schedule {v_j}={0,1,0,0,0,0,0,0,0,0} because the former can reuse the network resource more effectively, even though both are valid interference-avoiding or collision-free scheduling techniques.

Therefore, to improve the network resource utilization, the present invention based on the network condition at the time, such as urgency and transmission rate of each network link, weights each valid local transmission pattern. Without the loss of generality, the present invention uses a single cell multihop wireless network as an embodiment for explanation. However, the weighting scheme is also applicable to a multi-cell multi-hop wireless network.

For agent node f_i, the information on the packet queue length of the neighboring agent nodes are collected, and the weigh of local transmission pattern of the k-th local link of each agent node f_i is computed as follows:

$\omega t\left(f_i,\left\{0,\dots ,0,1,0,\dots ,0\right\}\right)=\mathrm{sum}\mathrm{of}\left\{\max \left(x_{T,t}\left(k\right)-x_{R,t}\left(k\right),0\right)/\max \left(x_{T,t}\left(h\right)-x_{R,t}\left(h\right),\varepsilon \right)\right\},$

where max is the maximum function, ε is a positive number, h is not equal to k, f_h belongs to the set of all the variable nodes linking to the agent node f_i, and x_T,t(k) and x_R,t(k) are the total number of packets that can reach the destination queue through the k-th local ink of agent node f_i, where subscripts T and R are the transmitting end and the receiving end of the local link, and t is the concerned packet slot index.

FIG. 4A is the factor graph of FIG. 3A with information on the packet queue length of each network link, using downlink for communication. The weight of local transmission pattern {v₁, v₂, v₃, v₄} of BS (agent node f₁) is computed as follows:

${\omega }_t\left(f_1,\left\{1,0,0,0\right\}\right)=\left(7+4+6-\left(5+6+3\right)\right)/6+\left(7+4+6-\left(5+6+3\right)\right)/\left(4+6+5-\left(3+5+4\right)\right)+\left(7+4+6-\left(5+6+3\right)\right)/5=3/6+3/3+3/5=2.6;$ ${\omega }_t\left(f_1,\left\{0,1,0,0\right\}\right)=6/\left(17-14\right)+6/\left(15-12\right)+6/5=5.2;$ ${\omega }_t\left(f_1,\left\{0,0,1,0\right\}\right)=\left(15-12\right)/\left(17-14\right)+\left(15-12\right)/6+\left(15-12\right)/5=2.1;$ ${\omega }_t\left(f_1,\left\{0,0,0,1\right\}\right)=5/\left(17-14\right)+5/6+5/\left(15-12\right)=2.1.$

The weighting of relay is the same as the weighting of BS, except that the weigh of the link that is in use and is the only link to the BS is set to be 1. Therefore, the weight of local transmission pattern {v₁,v₅,v₆,v₈} corresponding to agent node f₂ is computed as follows: ω_t(f₂, {1,0,0,0})=1; ω_t(f₂, {0,1,0,0})=5/6+5/3=2.5; ω_t(f₂, {0,0,1,0})=6/5+6/3=3.2; and ω_t(f₂, {0,0,0,1})=3/5+3/6=1.1.

The weighting of the uplinks is the same as the weighting of the downlinks, except that the transmission direction is reverse, and the packet destination is BS. FIG. 4B is the factor graph of FIG. 3A with information on the packet queue length of each network link, using uplink for communication. The weigh of local transmission pattern {v₁, v₂, v₃, v₄} of BS (agent node f₁) is computed as follows:

ω_t(f₁, {1,0,0,0})=7/8+7/6+7/7=3.0;

ω_t(f₁, {0,1,0,0})=8/7+8/6+8/7=3.6;

ω_t(f₁, {0,0,1,0})=6/7+6/8+6/7=2.5;

ω_t(f₁, {0,0,0,1})=7/7+7/8+7/6=3.0.

The weight of non-zero local transmission pattern {v₁, v₅, v₆, v₈} of agent node f₂ is computed as follows: ω_t(f₂, {1,0,0,0})=1; ω_t(f₂, {0,1,0,0})=(9−7)/ε+(9−7)/(8−7)=2(ε+1); ω_t(f₂, {0,0,1,0})=0/(9−7)=0; and ω_t(f₂, {0,0,0,1})=(8−7)/(9−7)+(8−7)/ε=ε+0.5.

The weight of non-zero local transmission pattern {v₂, v₈, v₉} of agent node f₂ is computed as follows: ω_t(f₆, {1,0,0})=ω_t(f₆, {1,0,0})=ω_t(f₆, {1,0,0})=1.

As shown in the above description, the present invention does not need to estimate the SNR. It is only necessary for the neighboring wireless APs to reach a consensus. The weighting scheme of the present invention is natural-competition-based. When the packet queue length of the transmitting end is longer than the packet queue length of the receiving end, the corresponding network link is more likely to be activated for use because the weigh of transmission pattern activating the network link is increased. This characteristic allows the urgent packets having a higher priority, and prevents the packet queue from overflowing on any network node. On the other hand, when the packet queue length of the receiving end is longer than the packet queue length of the transmitting end, the corresponding network link is more likely to be deactivated because the weight of transmission pattern activating the network link is set to 0. Stopping the network link prevents the packet queue from overflowing on any network node, and gives more time for the relay to consume the packets.

It is worth noticing that weighting scheme of the present invention is also applicable to a multi-cell multihop wireless network.

In accordance with the above fully-distributed packet scheduling method for a wireless network, FIG. 5 shows a block diagram of the apparatus realizing the fully-distributed packet scheduling method. As shown in FIG. 5, the apparatus includes a network modeling unit 501, a packet scheduling unit 503 and a weighting mechanism 505. Network modeling unit 501 is to model a multi-hop wireless network with a factor graph. As aforementioned, the factor graph includes a plurality of agent nodes, a plurality of variable nodes and a plurality of edges. The factor graph is bounded by a group of mutually-interactive constrain rules. Based on the factor graph, packet scheduling unit 503 transforms the packet scheduling problem into a decoding problem of LDPC code, and uses a standard process to solve the decoding problem. Weighting mechanism 505 weights the packet schedule solved by the decoding problem according to the network condition at the time.

The correspondence between the factor graph and the wireless network is described earlier in the construction of factor graph, and therefore, is omitted here.

The present invention is also compared with two conventional packet scheduling techniques, namely, round-robin (RR) and individual-polling (IP). RR and IP are both central-unit-processing-based packet scheduling methods. In RR, all the links of the network are activated in turn with a pre-determined order, while the local link between BS and relay is activated in an order that is determined individually in IP.

FIG. 6 shows the comparison between the present and the RR and IP in terms of average transmission rate in each packet slot, and the uplinks are used for communication in a single cell multihop wireless network. The x-axis is the mean of packet arrival rates λ, and the y-axis is the average of transmission rates in each packet slot.

As shown in FIG. 6, when the traffic load in the network is less than the capacity of the system (i.e., λ≦0.5), the weighting scheme of the present invention helps to maintain the fairness of packet transmission rate among all the mobile stations, regardless of the network resources used by each mobile station. Because of the fairness of transmission pattern, the priority of the urgent packets is raised. The fairness of packet scheduling can be seen in the standard deviation (STD) at upper left corner of the table.

When the traffic load in the network is gradually greater than the capacity of the system (i.e., λ≧0.5), the present invention releases the network resource of M1 to help other mobile stations to survive under the heavier traffic load.

Therefore, the average transmission rate in each packet slot and the difference among the users are far better than the two conventional packet scheduling methods.

FIG. 7 shows the average transmission rates of M1-M8 in each packet slot with uplinks for communication in a multi-cell multi-hop wireless network. The x-axis is the mean of packet arrival rates λ, and the y-axis is the average of transmission rates in each packet slot. Similarly, the weighting scheme helps to maintain the fairness of packet transmission rate among all the mobile stations, as well as helps other mobile stations to survive under the heavier traffic load.

As shown in FIG. 7, the network resources available to each user is not only related to the number of links available, but also related to the traffic consumption service surrounding the user. Although in a multi-cell multihop wireless network, the traffic information among neighboring nodes is complicated, the fully-distributed packet scheduling of the present invention is natural-competition-based. Therefore, regardless of the average packet arrival rate λ, the natural-competition-based method provides the optimal balanced solution.

Although the present invention has been described with reference to the preferred embodiments, it will be understood that the invention is not limited to the details described thereof. Various substitutions and modifications have been suggested in the foregoing description, and others will occur to those of ordinary skill in the art. Therefore, all such substitutions and modifications are intended to be embraced within the scope of the invention as defined in the appended claims.

Claims

1. A method of full-distributed packet scheduling for a wireless network, comprising the steps of: constructing a factor graph to model a multi-hop wireless network, said factor graph including a plurality of agent nodes, a plurality of variable nodes and a plurality of edges, said factor graph being a graph of a group of mutually-interactive constrain rules;transforming said packet scheduling problem into a decoding problem for LDPC code based on said factor graph, and solving said decoding problem through a standard process; andapplying a weighting scheme on the packet schedule solved on said decoding problem according to the network condition at the time.

2. The method as claimed in claim 1, wherein said wireless network is a multihop wireless network.

3. The method as claimed in claim 1, wherein said step of constructing said factor graph further includes the steps of: replacing each network node by an agent node represent, each said agent node marking with a corresponding constrain function fi, which defining an interference-avoiding local constrain rule;replacing each network link by a variable node, each said variable node marking with a variable vj; andlinking each said variable node to two said agent nodes, wherein said network nodes corresponding to two said agent nodes are able to communicate through the network link corresponding to said variable node.

4. The method as claimed in claim 1, wherein said wireless network is a relay network.

5. The method as claimed in claim 1, wherein each said network node obeys a interference-avoiding local constrain rule in each packet slot.

6. The method as claimed in claim 5, wherein said interference-avoiding local constrain rule includes the following constrain rules: a network node only transmits to one network node during transmission;a relay does not transmit and receive packets simultaneously; anda network node does not receive packets from multiple sources simultaneously.

7. The method as claimed in claim 1, wherein said packet scheduling further includes the computation and the transportation of a soft-information of each said variable node to improve network resource utilization, and said soft-information indicates the probability that a network link to be activated in each packet slot.

8. The method as claimed in claim 7, wherein said standard process computes said soft-information of said variable node through a sum-product algorithm.

9. The method as claimed in claim 1, wherein said network condition at the time depends on the urgency of each packet and the transmission rate of each said network link.

10. The method as claimed in claim 1, wherein said network condition at the time depends on the packet queue length and the transmission rate of each said network link.

11. The method as claimed in claim 1, wherein said weighting scheme includes collecting the information on the packet queue length of the neighboring agent nodes for each said agent node fi, and computing the weigh of said agent node fi in the local transmission pattern of each local link of said agent node fi

12. The method as claimed in claim 1, wherein said weight of each said agent node fi in the local transmission pattern of the k-th local link of said agent node fi is computed as follows:

13. An apparatus of full-distributed packet scheduling for a wireless network, comprising: a network modeling unit, for modeling a multi-hop wireless network through a factor graph, said factor graph including a plurality of agent nodes, a plurality of variable nodes and a plurality of edges, said factor graph being a graph of a group of mutually-interactive constrain rules;a packet scheduling unit, for transforming said packet scheduling problem into a decoding problem for LDPC code based on said factor graph, and solving said decoding problem through a standard process; anda weighting mechanism for being applied on the packet schedule solved on said decoding problem according to the network condition at the time.

14. The apparatus as claimed in claim 13, wherein said wireless network is a multihop wireless network.

15. The apparatus as claimed in claim 13, wherein said wireless network is a relay network.

16. The apparatus as claimed in claim 13, wherein said network condition at the time depends on the urgency of each packet and the transmission rate of each said network link

17. The apparatus as claimed in claim 13, wherein said network condition at the time depends on the packet queue length and the transmission rate of each said network link.

18. The apparatus as claimed in claim 13, wherein the correspondence between said factor graph and said wireless network is: each said agent node representing a network node, each said agent node marked with a corresponding constrain function fi, which defining an interference-avoiding local constrain rule; andeach said variable node representing a network link, each said variable node marked with a corresponding variable vj;where each said variable node is linked to two said agent nodes, and said network nodes corresponding to two said agent nodes are able to communicate through the network link corresponding to said variable node.