Iterative interference cancellation using mixed feedback weights and stabilizing step sizes

Abstract

A receiver is configured for canceling intra-cell and inter-cell interference in coded, multiple-access, spread-spectrum transmissions that propagate through frequency-selective communication channels. The receiver employs iterative symbol-estimate weighting, subtractive cancellation with a stabilizing step-size, and mixed-decision symbol estimates. Receiver embodiments may be implemented explicitly in software or programmed hardware, or implicitly in standard Rake-based hardware either within the Rake (i.e., at the finger level) or outside the Rake (i.e., at the user or subchannel symbol level).

Description

BACKGROUND

1. Field of the Invention

The present invention relates generally to iterative interference cancellation in received wireless communication signals and, more particularly, to cancellation of intra-cell interference and/or inter-cell interference in coded spread spectrum communication systems.

2. Discussion of the Related Art

In an exemplary wireless multiple-access system, a communication resource is divided into code-space subchannels that are allocated to different users. A plurality of subchannel signals received by a wireless terminal (e.g., a subscriber unit or a base station) may correspond to different users and/or different subchannels allocated to a particular user.

If a single transmitter broadcasts different messages to different receivers, such as a base station in a wireless communication system broadcasting to a plurality of mobile terminals, the channel resource is subdivided in order to distinguish between messages intended for each mobile. Thus, each mobile terminal, by knowing its allocated subchannel(s), may decode messages intended for it from the superposition of received signals. Similarly, a base station typically separates received signals into subchannels in order to differentiate between users.

In a multipath environment, received signals are superpositions of time-delayed and complex-scaled versions of the transmitted signals. Multipath can cause several types of interference. Intra-channel interference occurs when the multipath time-delays cause subchannels to leak into other subchannels. For example, in a forward link, subchannels that are orthogonal at the transmitter may not be orthogonal at the receiver. When multiple base stations (or sectors or cells) are active, there may also be inter-channel interference caused by unwanted signals received from other base stations. Each of these types of interference can degrade communications by causing a receiver to incorrectly decode received transmissions, thus increasing a receiver's error floor. Interference may also have other deleterious effects on communications. For example, interference may lower capacity in a communication system, decrease the region of coverage, and/or decrease maximum data rates. For these reasons, a reduction in interference can improve reception of selected signals while addressing the aforementioned limitations due to interference.

These interferences take the following form when code division multiplexing is employed for a communication link, either with code division multiple access (as used in CDMA 2000, WCDMA, and related standards) or with time division multiple access (as used in EV-DO and related standards). A set of symbols is sent across a common time-frequency slot of the physical channel and separated using a set of distinct code waveforms, which are usually chosen to be orthogonal (or pseudo-orthogonal for reverse-link transmissions). The code waveforms typically vary in time, and these variations are introduced by a pseudo-random spreading code (PN sequence). The wireless transmission medium is characterized by a time-varying multipath profile that causes multiple time-delayed replicas of the transmitted waveform to be received, each replica having a distinct amplitude and phase due to path loss, absorption, and other propagation effects. As a result, the received code set is no longer orthogonal. The code space suffers from intra-channel interference within a base station as well as inter-channel interference arising from transmissions in adjacent cells.

The most basic receiver architecture employed to combat these various effects is the well-known Rake receiver. The Rake receiver uses a channel-tracking algorithm to resolve the received signal energy onto various multipath delays. These delayed signals are then weighted by the associated complex channel gains (which may be normalized by path noise powers) and summed to form a single resolved signal, which exploits some of the path diversity available from the multipath channel. It is well known that the Rake receiver suffers from a significant interference floor, which is due to both self-interference from the base station of interest (or base stations, when the mobile is in a soft-handoff base station diversity mode) and multiple-access interference from all base stations in the coverage area. This interference limits the maximum data rates achievable by the mobiles within a cell and the number of mobiles that can be supported in the cell.

Advanced receivers have been proposed to overcome the limitations of the Rake receiver. The optimal multi-user detector (MUD) has the best performance, but is generally too computationally complex to implement. MUD complexity increases exponentially with respect to the total number of active subchannels across the cell of interest and the interfering cells as well as the constellation size(s) of the subchannels. This complexity is so prohibitive that even efficient implementations based on the Viterbi algorithm cannot make it manageable in current hardware structures. Another approach is a properly designed linear receiver, which in many channel scenarios, is able to retain much of the optimal MUD performance, but with a complexity that is polynomial in the number of subchannels. The most common examples are the linear minimum mean squared error (LMMSE) receiver and the related decorrelating (or zero-forcing) receiver, which both require finding, or approximating, the inverse of a square matrix whose dimension is equal to the lesser between the number of active subchannels and the length (in samples) of the longest spreading code.

Complexity can still be prohibitive with these receivers, because such a matrix inverse needs to be calculated (or approximated) for each symbol. These receivers depend not only on the spectral characteristics of the multipath fading channel (which could be slowly time varying), but also on the time-varying spreading codes employed on the subchannels over each symbol. Thus, these receivers vary at the symbol rate even if the channel varies much more slowly.

An alternative approach currently under development for advance receivers sidesteps the need to invert a matrix for each symbol. It accomplishes this by employing a PN-averaged LMMSE (PNA-LMMSE) receiver that assumes the PN code is random and unknown at the receiver (at least for determining the correlation matrix). While this receiver is generally inferior to the LMMSE approach, it has the advantage of not having to be implemented directly, because it is amenable to adaptive (or partially adaptive) implementations. The advantages of an adaptive implementation over a direct implementation include reduced complexity and the fact that the additive noise power (i.e., background RF radiation specific to the link environment, noise in the receiver's RF front end, and any processing noise such as noise due to quantization and imperfect filtering) does not have to be estimated. However, these advantages incur the costs associated with adaptive filters (e.g., performance and adaptation rate). Note that a direct implementation without knowledge of the noise power modifies the LMMSE and PNA-LMMSE receivers into the corresponding decorrelating (or zero-forcing) receivers that arise from taking the background noise power to be zero when deriving the LMMSE and PNA-MMSE receivers.

Another method for further reducing complexity is to iteratively approximate the matrix-inverse functionality of the LMMSE receiver without explicitly calculating the inverse. Receivers of this type employ multistage interference cancellation. One particular type is known as parallel interference cancellation (PIC), and is motivated by well-known iterative techniques of quadratic minimization. In each stage of PIC, the data symbols of the subchannels are estimated. For each subchannel, an interference signal from the other subchannels is synthesized, followed by interference cancellation that subtracts the synthesized interference from each subchannel. The interference-cancelled subchannels are then fed to a subsequent PIC stage. Ideally, within just a few stages (i.e., before the complexity grows too large), the performance rivals that of the full linear receiver using a matrix inverse.

PIC can be implemented in various modes depending on what types of symbol estimates are used for interference cancellation. In a soft-cancellation mode, PIC does not exploit additional information inherent in the finite size of user constellations. That is, estimates of data symbols are not quantized to a constellation point when constructing interference signals. However, in some multiple-access schemes, the user constellations may be known (e.g., in an EV-DO link or in a WCDMA link without HSDPA users) or determined through a modulation classifier. In such cases, it is possible for PIC to be implemented in a hard-cancellation mode. That is, estimates of data symbols are quantized to constellation points (i.e., hard decisions) when constructing the interference signal.

In a mixed-cancellation mode, PIC employs a soft decision on each symbol whose constellation is unknown, and either a soft or hard decision on each symbol whose constellation is known, depending on how close the soft estimate is to the hard decision. Such a mixed-decision PIC typically outperforms both the soft-decision PIC and the hard-decision PIC. Moreover, it can also substantially outperform the optimal LMMSE receiver and promises even greater performance gains over PNA-LMMSE approaches currently under development for advanced receivers. The performance of soft-decision PIC is bounded by the optimal LMMSE.

SUMMARY OF THE INVENTION

In view of the foregoing background, embodiments of the present invention may provide a generalized interference-canceling receiver for canceling intra-channel and inter-channel interference in coded, multiple-access, spread-spectrum transmissions that propagate through frequency-selective communication channels. Receiver embodiments may employ a designed and/or adapted soft-weighting subtractive cancellation with a stabilizing step-size and a mixed-decision symbol estimator. Receiver embodiments may be designed, adapted, and implemented explicitly in software or programmed hardware, or implicitly in standard Rake-based hardware, either within the Rake (i.e., at the finger level) or outside the Rake (i.e., at the subchannel symbol level). Embodiments of the invention may be employed in user equipment on the forward link and/or in a base station on the reverse link.

Some embodiments of the invention address the complexity of the LMMSE approach by using a low-complexity iterative algorithm. Some embodiments of the invention in soft-mode may be configured to achieve LMMSE performance (as contrasted to the lesser-performing PNA-LMMSE) using only quantities that are easily measured at the receiver. Some embodiments address the sub-optimality of the LMMSE and PNA-LMMSE approaches by using an appropriately designed mixed-decision mode and may even approach the performance of an optimal multi-user detector. In some embodiments, stabilizing step sizes may be used to enhance stability of various PIC approaches. Some embodiments may employ symbol-estimate weighting to control convergence of various PIC approaches. Some embodiments of the invention address the limitation of various PIC approaches to binary and quaternary phase shift keying in mixed-decision mode by being configurable to any subchannel constellation. Some embodiments of the invention address the difficulty of efficiently implementing various PIC approaches in hardware by using a modified Rake architecture. Some embodiments of the invention address the so-called “ping-pong effect” (i.e., when the symbol error rate oscillates with iteration) in various PIC approaches by pre-processing with a de-biasing operation when making symbol estimates.

In one embodiment of the invention, an iterative interference canceller comprises a weighting means configured for applying at least one symbol weight to the input symbol decisions, a stabilizing step size means configured for applying a stabilizing step size to an error signal, and a mixed-decision processing means. The canceller is iterative, and thus, the weighting means, the stabilizing step size means, and the mixed-decision processing means are configured to perform processing during each of a plurality of iterations for each of the input symbol decisions.

The mixed-decision processing means may include, by way of example, but without limitation, a combination of hardware and software configured to produce soft and/or hard symbol estimates. The mixed-decision means comprises a de-biasing means configured for scaling the input symbol estimates with a scale factor to remove bias computed on the input symbol estimates, and a processing means configured for processing each de-biased input symbol estimate, irrespective of other symbol estimates. The processing means produces a hard decision that quantizes the de-biased input symbol estimate onto a nearby constellation point, or a soft decision that scales the de-biased input symbol estimate.

The stabilizing step size means may include, by way of example, but without limitation, any combination of hardware and software configured to scale an error signal with a scaling factor that may be used for controlling convergence in an iterative canceller. For example, the stabilizing step size means may include a signal processor configured to calculate at least one stabilizing step size and a multiplier for scaling an error signal with the step size. The multiplier means is configured for scaling (e.g., multiplying) an error signal with the stabilizing step size.

The step size calculation means may include, by way of example, but without limitation, software or programmable hardware configured for calculating a stabilizing step size.

The weighting means may include, by way of example, but without limitation, a weight-calculation means configured for producing symbol weights, and a multiplier configured for scaling symbol estimates by the weights.

Embodiments of the invention may be employed in any receiver configured to support one or more CDMA standards, such as (1) the “TIA/EIA-95-B Mobile Station-Base Station Compatibility Standard for Dual-Mode Wideband Spread Spectrum Cellular System” (the IS-95 standard), (2) the “TIA/EIA-98-C Recommended Minimum Standard for Dual-Mode Wideband Spread Spectrum Cellular Mobile Station” (the IS-98 standard), (3) the standard offered by a consortium named “3rd Generation Partnership Project” (3GPP) and embodied in a set of documents including Document Nos. 3G TS 25.211, 3G TS 25.212, 3G TS 25.213, and 3G TS 25.214 (the WCDMA standard), (4) the standard offered by a consortium named “3rd Generation Partnership Project 2” (3GPP2) and embodied in a set of documents including “TR-45.5 Physical Layer Standard for cdma2000 Spread Spectrum Systems,” the “C.S0005-A Upper Layer (Layer 3) Signaling Standard for cdma2000 Spread Spectrum Systems,” and the “C.S0024 CDMA2000 High Rate Packet Data Air Interface Specification” (the CDMA2000 standard), (5) Multi-Code CDMA systems, such as High-Speed-Downlink-Packet-Access (HSDPA), and (6) other CDMA standards.

Receivers and cancellation systems described herein may be employed in subscriber-side devices (e.g., cellular handsets, wireless modems, and consumer premises equipment) and/or server-side devices (e.g., cellular base stations, wireless access points, wireless routers, wireless relays, and repeaters). Chipsets for subscriber-side and/or server-side devices may be configured to perform at least some of the receiver and/or cancellation functionality of the embodiments described herein.

These and other embodiments of the invention are described with respect to the figures and the following description of the preferred embodiments.

BRIEF DESCRIPTION OF THE DRAWINGS

Embodiments according to the present invention are understood with reference to the schematic block diagrams of FIGS. 1 through 12.

FIG. 1 is a general schematic illustrating an iterative interference canceller.

FIG. 2 is a block diagram illustrating a front-end processor for an iterative interference canceller.

FIG. 3 is a general schematic illustrating an interference cancellation unit (ICU).

FIG. 4 shows a weighting block in an ICU configured to separately process input symbol estimates corresponding to a plurality of base stations.

FIG. 5A is a block diagram illustrating part of an interference cancellation unit configured to synthesize constituent finger signals.

FIG. 5B is a block diagram illustrating part of an interference cancellation unit configured to synthesize constituent user signals.

FIG. 6A shows a cancellation block configured to perform interference cancellation on constituent signals, followed by Rake processing and despreading.

FIG. 6B shows a cancellation block configured to cancel interference in constituent signals, preceded by Rake processing and despreading.

FIG. 7 is a block diagram of the interference cancellation part of a subtractive canceller in which cancellation occurs prior to signal despreading.

FIG. 8A is a block diagram illustrating post interference-cancellation signal despreading on constituent finger signals.

FIG. 8B is a block diagram illustrating post interference-cancellation signal despreading on constituent user signals.

FIG. 9A is a block diagram showing a method for implicitly despreading a signal in a subtractive canceller that performs interference-cancellation prior to signal despreading.

FIG. 9B is a block diagram showing a method for explicitly despreading a signal in a subtractive canceller that performs interference-cancellation prior to signal despreading

FIG. 10 is a block diagram of a subtractive canceller configured to perform interference cancellation prior to signal despreading.

FIG. 11A is a block diagram illustrating an embodiment for implicitly calculating a stabilizing step size.

FIG. 11B is a block diagram illustrating how linear functions, such as despreading and generating a difference signal, can be swapped in an alternative embodiment for calculating a stabilizing step size.

FIG. 11C is a block diagram illustrating another embodiment for implicitly calculating a stabilizing step size.

FIG. 12 is a block diagram of a symbol-estimation block in an interference cancellation unit.

FIG. 13 is a block diagram of a dual feedback algorithm configured for implementing an iterative interference canceller.

Various functional elements or steps, separately or in combination, depicted in the figures may take the form of a microprocessor, digital signal processor, application specific integrated circuit, field programmable gate array, or other logic circuitry programmed or otherwise configured to operate as described herein. Accordingly, embodiments may take the form of programmable features executed by a common processor or discrete hardware unit.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

The present invention will now be described more fully hereinafter with reference to the accompanying drawings, in which preferred embodiments of the invention are shown. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

First the invention will be described as it applies to a forward-link channel, and then extended to include reverse-link channels. The following formula represents an analog baseband signal received at a mobile from multiple base stations, each with its own multipath channel,

$\begin{array}{cc}y\left(t\right)=\sum _{s=0}^{B-1}\sum _{l=0}^{L_{\left(s\right)}}{\alpha }_{\left(s\right),l}\sum _{k=0}^{K_{\left(s\right)}-1}{b}_{\left(s\right),k}{u}_{\left(s\right),k}\left(t-{\tau }_{\left(s\right),l}\right)+w\left(t\right),t\in \left(0,T\right),& \mathrm{Equation}1\end{array}$ with the following definitions

• • (0, T) is the symbol interval;
  • B is the number of modeled base stations and is indexed by the subscript (s) which ranges from (0) to (B−1); here, and in the sequel, the term “base stations” will be employed loosely to include cells or sectors;
  • L_(s) is the number of resolvable (or modeled) paths from base station (s) to the mobile;
  • α_(s),l and τ_(s),l are the complex gain and delay, respectively, associated with the l-th path of base station (s);
  • K_(s) is the number of active users or subchannels in base station (s) that share a channel via code-division multiplexing; these users or subchannels are indexed from 0 to K_(s)−1;
  • u_(s),k(t) is a code waveform (e.g., spreading waveform) of base station (s) used to carry the k^th user's symbol for that base station (e.g., a chip waveform modulated by a user-specific Walsh code and covered with a base-station specific PN cover);
  • b_(s),k is a complex symbol transmitted for the k^th user or subchannel of base station (s);
  • and w(t) is zero-mean complex additive noise that contains both thermal noise and any interference whose structure is not explicitly modeled (e.g., inter-channel interference from unmodeled base stations and/or intra-channel interference from unmodeled paths).

Typically, a user terminal (e.g., a handset) is configured to detect only symbols transmitted from its serving base station (e.g., the symbols from base station (0)) or a subset thereof (e.g., symbols for the k^th user of base station (0)). Interference can impede the determination of b_(s),k from y(t). Not only is additive noise w(t) present, but there may be intra-channel and inter-channel interference.

Intra-channel interference typically occurs when multiple users are served by a given base station (i.e., a serving base station). Even if the users' transmitted code waveforms are orthogonal, multipath in the transmission channel causes the codes to lose their orthogonality. Inter-channel interference is caused by transmissions from non-serving base stations whose signals contribute to the received baseband signal y(t).

FIG. 1 is a block diagram of an iterative interference canceller (IIC), which is a low-complexity receiver configured to mitigate intra-channel and inter-channel interference. The received baseband signal y(t) is input to a front-end processor 101, which produces initial symbol estimates for all symbols of the active users served by at least one base station. The initial symbol estimates are coupled to a first interference cancellation unit (ICU) 102 configured to cancel a portion of the intra-channel and inter-channel interference that corrupts the symbol estimates. The ICU 102 outputs a first set of updated symbol estimates, which are interference-cancelled symbol estimates. The updated symbol estimates are coupled to a second ICU 103. A plurality M of ICUs 102-104 illustrate an iterative process for performing interference cancellation in which the initial symbol estimates are updated M times.

FIG. 2 is a block diagram of the front-end processor 101 shown in FIG. 1. Each of a plurality B of Rake-based receiver components 201-203 provides estimates of symbols transmitted from a corresponding base station. The detailed block diagram depicted in Rake receiver 202 represents the functionality of each of the components 201-203. Rake receiver 202, corresponding to an s^th base station 202, includes a plurality L_(s) of delay elements 210-211 configured to advance the received baseband signal y(t) in accordance with multipath-delay quantities {τ_(s),l}_l=0^L(s)−1. The advanced signals are scaled 212-213 by corresponding path gains {α_(s),l}_l=0^L(s)−1 prior to combining 214 to produce a combined signal of the form

$\sum _{l=0}^{L_{\left(s\right)}-1}\frac{{\alpha }_{\left(s\right),l}^{*}}{{\underset{\_}{\alpha }}_{\left(s\right)}}y\left(t+{\tau }_{\left(s\right),l}\right),$ where

${\underset{\_}{\alpha }}_{\left(s\right)}={\left(\sum _{l=0}^{L_{\left(s\right)}-1}{{\alpha }_{\left(s\right),l}}^2\right)}^{1/2}$ is the Euclidean norm of the path-gain vector,α_(s)=└α_(s),0 α_(s),1 . . . α_(s),L(s)−1┘^T, and the superscript T denotes the matrix transpose operator.

The combined signal is resolved onto the users' code waveforms by correlative despreading, which comprises multiplying 215-216 the combined signals by complex conjugates of each code waveform, followed by integrating 217-218 the resultant products. A despread signal corresponding to a k^th code waveform is

$\begin{array}{cc}{q}_{\left(s\right),k}\equiv \frac{1}{{{\underset{\_}{\alpha }}_{\left(s\right)}}^2}{\int }_0^T{u}_{\left(s\right),k}^{*}\left(t\right)\sum _{l=0}^{L_{\left(s\right)}-1}{\alpha }_{\left(s\right),l}^{*}y\left(t+{\tau }_{\left(s\right),l}\right)dt.& \mathrm{Equation}2\end{array}$

This value is also referred to as a Rake front-end soft estimate of the symbol b_(s),k. Since Rake processing, combining, and despreading are linear operations, their order may be interchanged. Thus, alternative embodiments may be provided in which the order of the linear operations is changed to produce q_(s),k.

A symbol estimator comprises scaling blocks 219-220 and function blocks 221-222, which are configured to refine the estimates q_(s),k into front-end symbol estimates {circumflex over (b)}_(s),k^[0] of the transmitted data symbols b_(s),k. Each of the functions depicted in FIG. 2 may be configured to process discrete-time sequences. For example, time advances 210-211 (or delays) may be implemented as shifts by an integer number of samples in discrete-time sequences, and integration 217-218 may employ summation.

FIG. 3 is a block diagram of an i^th ICU comprising four functional blocks. A weighting module 301 calculates and applies soft weights to input symbol estimates. A synthesizing module 302 processes weighted symbol estimates to synthesize constituent signals of an estimated received signal. For example, the estimated received signal y(t) is a sum of the constituent signals, each of which is synthesized from the weighted symbol estimates. The synthesized constituents are processed in a canceller 303 (such as a subtraction module) configured to produce interference-cancelled signals having reduced intra-channel and inter-channel interferences. The canceller 303 also includes a resolving module (not shown) configured to resolve the interference-cancelled signals onto user code waveforms to produce resolved signals. A mixed-decision module 304 processes the resolved signals to produce updated symbol estimates.

FIG. 4 shows a weighting module (such as weighting module 301) configured to separately process input symbol estimates corresponding to a plurality B of base stations. A plurality of scaling modules 401-403 scale the input symbol estimates. Scaling module 402 depicts detailed functionality for processing signals from an exemplary s^th base station. Similar details are typically present in each of the scaling modules 401-403.

A plurality K_(s) of symbol estimates {{circumflex over (b)}_(s),k^[1]}_k=0^K(s)−1 of transmitted symbols {b_(s),k}_k=0^K(s)−1 produced by an i^th ICU is input to scaling module 402. The symbol estimates are multiplied 410-411 by corresponding complex weights {γ_(s),k^[i]}_k=0^K(s)−1 to produce weighted symbol estimates {γ_(s),k^[i]{circumflex over (b)}_(s),k^[i]}_k=0^K(s)−1. The magnitude of weight γ_(s),k^[i] may be calculated with respect to a merit of the corresponding symbol estimate {circumflex over (b)}_(s),k^[i].

The soft weights can be regarded as a confidence measure related to the accuracy of a decision, or symbol estimate. For example, a high confidence weight relates to a high certainty that a corresponding decision is accurate. A low confidence weight relates to a low certainty. Since the soft weights are used to scale decisions, low-valued weights reduce possible errors that may be introduced into a calculation that relies on symbol estimates.

In one embodiment of the invention, the weights γ_(s),k^[i] may be derived from at least one signal measurement, such as SINR. Clearly, the larger the SINR, the greater the reliability of the corresponding symbol estimate. For example, the weights γ_(s),k^[i] may be expressed by

$\begin{array}{cc}{\gamma }_{\left(s\right),k}^{\left[i\right]}=\max \left\{C_{\left(s\right),k},\frac{1}{1+1/{\mathrm{SINR}}_{\left(s\right),k}^{\left[i\right]}}\right\},& \mathrm{Equation}3\end{array}$ where SINR_(s),k^[i] is a ratio of average signal power to interference-plus-noise power of a k^th user in base station (s) after the i^th ICU, and C_k is a non-negative real constant that can be used to ensure some feedback of a symbol estimate, even if its SINR is small. Note that, as the SINR grows large, the weight tends toward unity, meaning that the estimate is very reliable.

The SINR (and thus, the soft weights) may be evaluated using techniques of statistical signal processing, including techniques based on an error-vector magnitude (EVM). Alternatively, a pilot-assisted estimate of the broadband interference-plus-noise floor, together with a user specific estimate of the signal-plus-interference-plus-noise floor, may be used to estimate the SINR values.

In another embodiment of the invention, the weights γ_(s),k^[i] may be expressed as a function of symbol estimates {circumflex over (b)}_(s),k^[i], such as shown in the following equation

$\begin{array}{cc}{\gamma }_{\left(s\right),k}^{\left[i\right]}=\frac{\mathrm{Re}\left\{E\left[{\mathrm{slice}\left({\hat{b}}_{\left(s\right),k}^{\left[i\right]}\right)}^{*}{\hat{b}}_{\left(s\right),k}^{\left[i\right]}\right]\right\}}{E\left[{{\hat{b}}_{\left(s\right),k}^{\left[i\right]}}^2\right]},& \mathrm{Equation}4\end{array}$ where Re{} returns the real part of the argument. The statistical expectations E[ ] in the numerator and denominator can be estimated, for example, via time-series averaging. The term slice({circumflex over (b)}_(s),k^[i]) represents the symbol estimate {circumflex over (b)}_(s),k^[i] sliced (i.e., quantized) to the nearest constellation point from which the symbol b_(s),k was drawn. This approach is applicable for symbols with known constellations. For example, it is typical for a receiver to know the symbol constellation for a user of interest, but it may not know which constellations are assigned to other users.

In this embodiment, the weights γ_(s),k^[i] are a function of a symbol estimate's {circumflex over (b)}_(s),k^[i] proximity to a given constellation point. Thus, a symbol estimate {circumflex over (b)}_(s),k^[i] that is close to a constellation point is provided with a large weight indicative of a high confidence measure in the symbol estimate's accuracy. For example, if the value {circumflex over (b)}_(s),k^[i] is a hard-decision estimate of b_(s),k (i.e., it is quantized to the nearest constellation point), then its associated weight is γ_(s),k^[i]=1, which indicates a high degree of confidence in the symbol estimate.

In some embodiments, both Equation 3 and Equation 4 may be used in a receiver to calculate soft weights. Some embodiments of the invention may provide for subset selection to force one or more of the weights to zero. Such embodiments may be expressed as adaptations to Equation 3 and/or Equation 4 expressed byγ_(s),k^[i]=0 for some subset of the users.  Equation 5

Forcing the weights of some users to zero effectively restricts which user signals are employed for interference cancellation. Some embodiments may provide for canceling only a predetermined number P of strongest users (e.g., users having the largest weight values). The number P may be fixed for all iterations, or it may vary with respect to iteration. In some embodiments, the number P may range from zero (i.e., no interference cancellation) to

$K\equiv \sum _{s=0}^{B-1}{K}_{\left(s\right)}$ (i.e., interference of all users cancelled). In some embodiments, the weights of user signals transmitted from at least one weakest base station are set to zero.

FIG. 5A is a block diagram of a synthesizing module (such as the synthesizing module 302) in which the constituent signals are associated with each Rake finger. Each of a plurality B of synthesizing modules 501-503 is assigned to one of a plurality B of base stations. A block diagram for an exemplary synthesizing module 502 corresponding to a base station (s) depicts details that are common to all of the synthesizing modules 501-503.

Weighted symbol estimates γ_(s),k^[i]{circumflex over (b)}_(s),k are modulated 510-511 onto corresponding code waveforms u_(s),k(t) to produce a plurality K_(s) of coded waveforms, which are combined in combining module 512 to produce a synthesized transmission signal

$\sum _{k=0}^{K_{\left(s\right)}-1}{\gamma }_{\left(s\right),k}^{\left[i\right]}{\hat{b}}_{\left(s\right),k}^{\left[i\right]}{u}_{\left(s\right),k}\left(t\right).$ Channel emulation (including delaying the synthesized transmission 513-514 by τ_(s),l and scaling 515-516 with channel gains α_(s),l) is performed to produce constituent signals corresponding to each finger. A synthesized constituent signal for an l^th finger of base station (s) is

$\begin{array}{cc}{\tilde{y}}_{\left(s\right),l}^{\left[i\right]}\equiv {\alpha }_{\left(s\right),l}\sum _{k=0}^{K_{\left(s\right)}-1}{\gamma }_{\left(s\right),k}^{\left[i\right]}{\hat{b}}_{\left(s\right),k}^{\left[i\right]}{u}_{\left(s\right),k}\left(t-{\tau }_{\left(s\right),l}\right).& \mathrm{Equation}6\end{array}$

When all of the finger constituents are summed, the result is

$\begin{array}{cc}{\tilde{y}}^{\left[i\right]}\left(t\right)\equiv \sum _{s=0}^{B-1}\sum _{l=0}^{L_{\left(s\right)}-1}{\tilde{y}}_{\left(s\right),l}^{\left[i\right]}\left(t\right),& \mathrm{Equation}7\end{array}$ which is an estimate of the signal that would be received at the mobile if the base stations were to transmit the weighted symbols.

FIG. 5B is a block diagram of a synthesizing module (such as the synthesizing module 302) in which the constituent signals are associated with each user in the system. Each synthesizing module 521-523 is configured to emulate multipath channels for all base stations. Synthesizing module 522 includes a block diagram that is indicative of the functionality of each of the synthesizing module 521-523

In synthesizing module 522, a plurality K_(s) of modulators (such as modulator 531) modulates each weighted symbol γ_(s),k^[i]{circumflex over (b)}_(s),k onto a corresponding code waveform u_(s),k(t). Each modulated code waveform is processed by a bank of finger delay elements 532-533 and channel gain scaling elements 534-535 corresponding to the multipath channel of base station (s). The resulting emulated multipath components are combined in combining module 536 to produce an estimated received signal for a k^th user of base station (s),

$\begin{array}{cc}{\tilde{y}}_{\left(s\right),k}^{\left[i\right]}\equiv \sum _{l=0}^{L_{\left(s\right)}-1}{\alpha }_{\left(s\right),l}{\gamma }_{\left(s\right),k}^{\left[i\right]}{\hat{b}}_{\left(s\right),k}^{\left[i\right]}{u}_{\left(s\right),k}\left(t-{\tau }_{\left(s\right),l}\right).& \mathrm{Equation}8\end{array}$

The subscript k on the left-hand side denotes that the constituent signal is for a user k, whereas the subscript l on the left-hand side of Equation 6 represents that the constituent signal is for a finger l. The sum of the user constituent signals produces a synthesized received signal

$\begin{array}{cc}{\tilde{y}}^{\left[i\right]}\left(t\right)\equiv \sum _{s=0}^{B-1}\sum _{k=0}^{K_{\left(s\right)}-1}{\tilde{y}}_{\left(s\right),k}^{\left[i\right]}\left(t\right).& \mathrm{Equation}9\end{array}$

The left-hand sides of Equation 7 and Equation 9 are the same signal, whereas the right-hand sides are simply two different decompositions.

FIG. 6A shows a cancellation module 601 (such as the canceller 303 in FIG. 3) configured to perform interference cancellation 610 on constituent signals, followed by Rake processing and despreading 611 in a Rake-based receiver. FIG. 6B shows a cancellation module 602 configured to synthesize 621 a received signal from constituent components, followed by Rake processing and despreading 622, and interference cancellation 623.

FIG. 7 is a block diagram of an interference canceller comprising a plurality B of cancellers 701-703 configured to perform interference cancellation on a plurality J of constituent signals for each of a plurality B of base stations. Since the constituent signals may be either fingers or users, index jε{0, 1, . . . , J_(s)−1} is expressed by

$J_{\left(s\right)}=\{\begin{array}{cc}{L}_{\left(s\right)}& \mathrm{for}\mathrm{finger}\mathrm{constituents}\\ K_{\left(s\right)}& \mathrm{for}\mathrm{user}\mathrm{constituents}\end{array}.$

Canceller 702 includes a block diagram that represents the functionality of each of the cancellers 701-703. The constituent signals corresponding to each base station are summed in a combining module 711 to produce a synthesized received signal,

${\tilde{y}}_{\left(s\right)}^{\left[i\right]}\equiv \sum _{j=0}^{J_{\left(s\right)}-1}{\tilde{y}}_{\left(s\right),j}^{\left[i\right]},$ where {tilde over (y)}_(s),j^[i] is j^th constituent signal (either finger or user) for base station (s). A plurality B of these sums corresponding to different base stations are combined in combining module 721 to produce a synthesized receive signal

${\tilde{y}}^{\left[i\right]}\left(t\right)=\sum _{s=0}^{B-1}{\tilde{y}}_{\left(s\right)}^{\left[i\right]}\left(t\right).$ The synthesized receive signal is subtracted from the actual received signal in a subtraction module 722 to produce a residual signal y(t)−{tilde over (y)}^[i](t). A stabilizing step size module 723 scales the residual signal by a complex stabilizing step size μ^[i] to produce a scaled residual signal μ^[i](y(t)−{tilde over (y)}^[i](t)). The scaled residual signal is combined with the constituent signals {tilde over (y)}_(s),j^[i] in combining modules 712-714 to produce a set of interference-cancelled constituents represented byz_(s),j^[i](t)≡{tilde over (y)}_(s),j^[i](t)+μ^[i](y(t)−{tilde over (y)}^[i](t)),  Equation 10where z_(s),j^[i](t) is an interference-cancelled j^th constituent signal for base station (s).

In an alternative embodiment, cancellation may be performed with only a subset of the constituent channels. In each base station, only those constituent signals being used for cancellation may be used to synthesize the estimated receive signal for base station (s). Thus, {tilde over (y)}_(s)^[i] becomes

${\tilde{y}}_{\left(s\right)}^{\left[i\right]}\equiv \sum _{j\in J_{\left(s\right)}}{\tilde{y}}_{\left(s\right),j}^{\left[i\right]}\mathrm{where}{J}_{\left(s\right)}\subseteq \left\{0,1,\dots ,J_{\left(s\right)}-1\right\}$ are indices of the subset of constituent signals to be employed in cancellation. Embodiments of the invention may be configured for applications in which hardware limitations restrict the number of finger signals or user signals that can be used for interference cancellation (e.g., only the strongest constituents are used).

The interference-cancelled signals produced by the canceller shown in FIG. 7 may be processed by a Rake despreader shown in FIG. 8A, which is configured for processing finger inputs. Specifically, finger signals associated with each base station are input to corresponding fingers of a Rake despreading module tuned to that base station. A Rake despreading module 802 tuned to an s^th base station comprises a block diagram indicating the functionality of a plurality B of Rake despreading modules 801-803.

Interference cancelled signals z_(s),l^[i](t) are time-advanced 810-811 by an amount τ_(s),l. A maximal ratio combining module scales 812-813 each time-advanced signal z_(s),l^[i](t+τ_(s),l) by α_(s),l*/∥α_(s)∥ and combines 814 the time-advanced signals for each base station. A resolving module comprising multipliers 815-816 and integrators 817-818 resolves each combined signal

$\frac{1}{{\underset{\_}{\alpha }}_{\left(s\right)}}\sum _{l=0}^{L_{\left(s\right)}-1}{\alpha }_{\left(s\right),k}^{*}{z}_{\left(s\right),l}^{\left[i\right]}\left(t+{\tau }_{\left(s\right),l}\right)$ onto code waveforms associated with base station (s) via correlative despreading. The resulting quantity for a k^th user of base station (s) is denoted by

$\begin{array}{cc}{\tilde{b}}_{\left(s\right),k}^{\left[i\right]}=\frac{1}{{{\underset{\_}{\alpha }}_{\left(s\right)}}^2}{\int }_0^T{u}_{\left(s\right),k}^{*}\left(t\right)\sum _{l=0}^{L_{\left(s\right)}-1}{\alpha }_{\left(s\right),k}^{*}{z}_{\left(s\right),l}^{\left[i\right]}\left(t+{\tau }_{\left(s\right),l}\right).& \mathrm{Equation}11\end{array}$

FIG. 8B is a block diagram of a Rake despreader configured for processing interference-cancelled signals relating to user inputs. In this case, the input constituent signals are user signals. A Rake despreading module 822 tuned to an s^th base station comprises a block diagram indicating functional details that are common to a plurality B of Rake despreading modules 821-823.

Interference cancelled signals z_(s),k^[i](t) corresponding to a k^th user and s^th base station are processed by a plurality L_(s) of time-advance modules 831-832 corresponding to the multipath channel for the s^th base station. The resulting time-advanced signals {z_(s),k^[i](t+τ_(s),l)}_l=0^L(s)−1 are weighted by a plurality of weighting modules 833-834, and the weighted signals are combined in combiner 835. A resolving module comprising multiplier 836 and integrator 837 resolves the combined signal onto the k^th user's code waveform to give

$\begin{array}{cc}{\tilde{b}}_{\left(s\right),k}^{\left[i\right]}=\frac{1}{{{\underset{\_}{\alpha }}_{\left(s\right)}}^2}{\int }_0^T{u}_{\left(s\right),k}^{*}\left(t\right)\sum _{l=0}^{L_{\left(s\right)}-1}{\alpha }_{\left(s\right),l}^{*}{z}_{\left(s\right),k}^{\left[i\right]}\left(t+{\tau }_{\left(s\right),l}\right).& \mathrm{Equation}12\end{array}$

The values of {tilde over (b)}_(s),k^[i] shown in Equation 11 and Equation 12 are generally not the same value, since the value of {tilde over (b)}_(s),k^[i] in Equation 11 is produced by cancellation employing finger constituents, whereas {tilde over (b)}_(s),k^[i] expressed by Equation 12 is produced by cancellation employing user constituents.

FIG. 9A is a block diagram of a Rake despreader, such as Rake despreaders 611 and 622 shown in FIGS. 6A and 6B, respectively. The Rake despreader comprises a plurality B of Rake despreading modules 901-903, each configured to process constituent signals from one of a plurality B of base stations. An exemplary Rake despreader module 902 is a block diagram illustrating functionality of each of the Rake despreader modules 901-903.

Input constituent signals {tilde over (y)}_(s),j^[i](t) for all values of j are subtracted 911 from the received signal y(t) to produce a difference signal, or error signal, representing the difference between the received signal and the synthesized estimates of signals received by the base stations. For base station (s), the difference signal is y(t)−{tilde over (y)}_(s)^[i](t), where

${\tilde{y}}_{\left(s\right)}^{\left[i\right]}\left(t\right)\equiv \sum _{j=0}^{J_{\left(s\right)}-1}{\tilde{y}}_{\left(s\right),j}^{\left[i\right]}\left(t\right).$ The difference signal for base station (s) is processed by a parallel bank of time advance modules 912-913 associated with the multipath channel for that base station, followed by maximal-ratio combining. In this embodiment, a maximal-ratio combining module is configured to perform weighting 914-915 and combining 916. A resolving module comprising multipliers 917-918 and integrators 919-920 resolves the resulting combined signals onto code waveforms of the base station's users to give the difference signal vector, or error signal vector, q_(s),k−{tilde over (q)}_(s),k^[i], where

$\begin{array}{cc}{\tilde{q}}_{\left(s\right),k}^{\left[i\right]}=\frac{1}{{{\underset{\_}{\alpha }}_{\left(s\right)}}^2}{\int }_0^T{u}_{\left(s\right),k}^{*}\left(t\right)\sum _{l=0}^{L_{\left(s\right)}-1}{\alpha }_{\left(s\right),l}^{*}{z}_{\left(s\right)}^{\left[i\right]}\left(t+{\tau }_{\left(s\right),l}\right)& \mathrm{Equation}13\end{array}$ and z_(s)^[i](t) was defined in Equation 10.

Rake despreading, such as described with respect to the exemplary Rake despreading module 902, may also be accomplished explicitly by employing matrix multiplication to synthesize constituent signals of the received signal, such as represented by block 931 shown in FIG. 9B.

A diagonal soft-weighting matrix may be defined asΓ^[i]=diag(γ_(0),0^[i], . . . , γ_(0),K(0)−1^[i]| . . . |γ_(B−1),0^[i], . . . , γ_{(B−1),K(B−1)−1}^[i]),  Equation 14in which all of the users' soft weights are ordered first by base station and then by users within a base station. The same indexing may also be used to express the column vector of symbol estimates input to an i^th ICU as{circumflex over (b)}^[i]=[{circumflex over (b)}_(0),0^[i], . . . , {circumflex over (b)}_(0),K(0)−1^[i]| . . . |{circumflex over (b)}_(B−1),0^[i], . . . , {circumflex over (b)}_{(B−1),K(B−1)−1}^[i]]^T.  Equation 15

The weighted symbol estimates are expressed as Γ^[i]b^[i], and the outputs of the Rake despreading modules 901-903 are expressed by the difference equation,q−{tilde over (q)}^[i]=q−RΓ^[i]b^[i],  Equation 16where,q=[q_(0),0, . . . , q_(0),K(0)−1| . . . |q_(B−1),0, . . . , q_{(B−1),K(B−1)−1}]^T  Equation 17and{tilde over (q)}^[i]=[{tilde over (q)}_(0),0^[i], . . . , {tilde over (q)}_(0),K(0)−1^[i]| . . . |{tilde over (q)}_(B−1),0^[i], . . . , {tilde over (q)}_{(B−1),K(B−1)−1}^[i]]^T.  Equation 18

The values of q^[i] represent the despread signals, such as described with respect to FIG. 2. The values of {tilde over (q)}^[i] are represented by Equation 13, and R is a square matrix whose elements are correlations between the users' received code waveforms. In FIG. 9B, the functionality expressed by Equation 16 is implemented via the matrix-multiplication block 931 and a subtraction module 932.

A global index κε{0, 1, . . . , K−1} with

$K\equiv \sum _{s=0}^{B-1}{K}_{\left(s\right)}$ is employed for ordering users (first by base station, and then by users within a base station) described with respect to Equation 14. Thus, κ=0 corresponds to a first user (denoted by index zero) of a first base station (denoted by an index zero), and κ=K corresponds to a last user (denoted by index K_(B−1)−1) of a last base station (denoted by index (B−1) ). If a user κ is a member of base station (s) and a user κ′ is a member of base station (s′), then the (κ, κ′) element of matrix R may be expressed by

$\begin{array}{cc}{R}_{{\mathrm{\kappa \kappa }}^{\prime }}=\frac{1}{{\underset{\_}{\alpha }}_{\left(s\right)}{\underset{\_}{\alpha }}_{\left(s^{\prime }\right)}}\int \sum _{l=0}^{L_{\left(s\right)}-1}{\alpha }_{\left(s\right),l}{u}_{\kappa }\left(t-{\tau }_{\left(s\right),l}\right)\sum _{l^{\prime }=0}^{L_{\left(s^{\prime }\right)}-1}{\alpha }_{\left(s^{\prime }\right),l^{\prime }}^{*}{u}_{{\kappa }^{\prime }}^{*}\left(t-{\tau }_{\left(s^{\prime }\right),l^{\prime }}\right)dt.& \mathrm{Equation}19\end{array}$

Thus, the elements of R can be built at the receiver with estimates of the path gains, path delays, and knowledge of the users' code waveforms.

FIG. 10 is a block diagram of an interference canceller, such as the interference-cancellation block 623 shown in FIG. 6. The difference signal q_(s),k−{tilde over (q)}_(s),k^[i] is scaled with a stabilizing step size μ⁽ⁱ⁾ by a stabilizing step size module 1001, which may include a calculation module (not shown) configured to calculate a stabilizing step size having a magnitude that is a function of proximity of input symbol decisions to a desired interference-cancelled symbol decision. The resulting scaled difference signal is summed 1003 with a product 1002 of the weighted symbol estimates Γ^[i]{circumflex over (b)}^[i] and a K×K implementation matrix F to yield{tilde over (b)}^[i+1]=FΓ^[i]{circumflex over (b)}^[i]+μ^[i](q−{tilde over (q)}^[i]).  Equation 20

The choice of F allows interference cancellation after despreading to mimic interference cancellation prior to despreading for either user constituents or finger constituents. For user constituents, F=I. For finger constituents, F is a block-diagonal matrix with a plurality B of diagonal blocks, wherein an s^th diagonal block is a K_(s)×K_(s) block representing the users' transmit correlation matrix for base station (s). The (k,k′) element of the s^th diagonal block (denoted by F_(s)(s)) is equal to(F_(s)(s))_kk′=ιu_(s),k(t)u_(s),k′*(t)dt.  Equation 21

The stabilizing step size μ^[i] may be used to enhance interference cancellation in each ICU and/or stabilize iterative interference cancellation. A quality metric of a canceller's output {tilde over (b)}^[i+1] may be derived as follows. If it is known (or approximated) that the additive noise w(t) in Equation 1 is Gaussian, then the despread outputs q, conditional on the transmitted symbolsb=[b_(0),0, . . . , b_(0),K(0)−1| . . . |b_(B−1),0, . . . , b_{(B−1),K(B−1)−1}]^T,  Equation 22are jointly complex normal random variables with mean Rb and covariance Γ^[i] R (i.e., q|b is distributed as CN(Rb; R)). If it is approximated that q|{tilde over (b)}^[i+1] is distributed as CN(R{tilde over (b)}^[i+1]; R), where {tilde over (b)}^[i+1] and its dependence on μ^[i] are given by Equation 20, then the value of μ^[i] that gives the maximum-likelihood soft estimate for {tilde over (b)}^[i+1] is

$\begin{array}{cc}{\mu }^{\left[i\right]}=\frac{{\left(\underset{\_}{q}-\mathrm{RF}{\Gamma }^{\left[i\right]}{\underset{\_}{\hat{b}}}^{\left[i\right]}\right)}^H\left(\underset{\_}{q}-R{\Gamma }^{\left[i\right]}{\underset{\_}{\hat{b}}}^{\left[i\right]}\right)}{{\left(\underset{\_}{q}-R{\Gamma }^{\left[i\right]}{\underset{\_}{\hat{b}}}^{\left[i\right]}\right)}^{H}R\left(\underset{\_}{q}-R{\Gamma }^{\left[i\right]}{\underset{\_}{\hat{b}}}^{\left[i\right]}\right)}.& \mathrm{Equation}23\end{array}$

Alternatively, the value of μ^[i] that gives the maximum-likelihood soft estimate for Γ^[i]{tilde over (b)}^[i+1] is

$\begin{array}{cc}{\mu }^{\left[i\right]}=\frac{{\left(\underset{\_}{q}-R{\Gamma }^{\left[i\right]}F{\Gamma }^{\left[i\right]}{\underset{\_}{\hat{b}}}^{\left[i\right]}\right)}^H{\Gamma }^{\left[i\right]}\left(\underset{\_}{q}-R{\Gamma }^{\left[i\right]}{\underset{\_}{\hat{b}}}^{\left[i\right]}\right)}{{\left(\underset{\_}{q}-R{\Gamma }^{\left[i\right]}{\underset{\_}{\hat{b}}}^{\left[i\right]}\right)}^H{\left({\Gamma }^{\left[i\right]}\right)}^{H}R{\Gamma }^{\left[i\right]}\left(\underset{\_}{q}-R{\Gamma }^{\left[i\right]}{\underset{\_}{\hat{b}}}^{\left[i\right]}\right)}.& \mathrm{Equation}24\end{array}$

Different formulations of the step-size may be used within the same IIC. For example, a step size based on Equation 24 may be used in a sequence of ICUs and Equation 23 may be used in the last ICU of the sequence. The Equations 23 and 24 may be adapted for cases in which there is no soft weighting (i.e., when Γ^[i]=I). Similarly step-size equations may be adapted when constituent user signals are employed (i.e., F=I). Furthermore, Equation 23 and Equation 24 may be determined implicitly whenever F=I, or when F is approximated as I. Since F contains the users' correlation matrices at the transmitter for each base station as its block diagonal, it will approximately equal identity, as the users' code waveforms are typically designed to be mutually orthogonal (or quasi-orthogonal for the reverse link). Any non-orthogonality is due to the finite duration of the pulse-shaping filters that approximate their infinite-duration theoretical counterparts. In this case, Equation 23 becomes

$\begin{array}{cc}{\mu }^{\left[i\right]}=\frac{{\left(\underset{\_}{q}-R{\Gamma }^{\left[i\right]}{\underset{\_}{\hat{b}}}^{\left[i\right]}\right)}^H\left(\underset{\_}{q}-R{\Gamma }^{\left[i\right]}{\widehat{\underset{\_}{b}}}^{\left[i\right]}\right)}{{\left(\underset{\_}{q}-R{\Gamma }^{\left[i\right]}{\widehat{\underset{\_}{b}}}^{\left[i\right]}\right)}^{H}R\left(\underset{\_}{q}-R{\Gamma }^{\left[i\right]}{\widehat{\underset{\_}{b}}}^{\left[i\right]}\right)}.& \mathrm{Equation}25\end{array}$

FIG. 11A illustrates a method and apparatus for calculating a stabilizing step size. A Rake receiver 1100 comprises a first Rake, maximal ratio combining, and despreading unit 1101 to process a received signal y(t) for producing an output despread signal vector q. A second Rake, maximal ratio combiner, and despreader unit 1102 processes a synthesized receive signal with weighted symbol estimates corresponding to an i^th iteration, and represented by

${\tilde{y}}^{\left[i\right]}\left(t\right)=\sum _{s=0}^{B-1}\sum _{l=0}^{L_{\left(s\right)}-1}{\alpha }_{\left(s\right),l}\sum _{k=0}^{K_{\left(s\right)}-1}{\gamma }_{\left(s\right),k}^{\left[i\right]}{\hat{b}}_{\left(s\right),k}^{\left[i\right]}{u}_{\left(s\right),k}\left(t-{\tau }_{\left(s\right),l}\right),$ to produce an estimated received signal RΓ^[i]{tilde over (b)}^[i].

A combiner 1103 calculates the difference between the outputs of 1101 and 1102 to produce a difference signal, or error signal, β^[i]≡q−RΓ^[i]b^[i], whose elements are indexed first by base station, and then by users within a base station,β^[i]=[β_(0),0^[i], . . . , β_(0),K(0)−1^[i]| . . . |β_(B−1),0^[i], . . . , β_{(B−1),K(B−1)−1}^[i]]^T.Alternatively, since the operations used to produce β^[i] are linear, a difference signal y(t)−{tilde over (y)}_(s)^[i](t) may be produced prior to despreading, such as shown by block 1110 in FIG. 11B.

The norm-square of β^[i] (i.e., ∥β^[i]∥²) is evaluated 1104 to generate the numerator in Equation 25. The elements of β^[i] are processed 1105 to produce a synthesized received signal

$\sum _{s=0}^{B-1}\sum _{l=0}^{L-1}{\alpha }_{\left(s\right),l}\sum _{k=0}^{K-1}{\beta }_{\left(s\right),k}^{\left[i\right]}{u}_{\left(s\right),k}\left(t-{\tau }_{\left(s\right),l}\right),$ and the norm square of this signal is calculated 1106 to produce the denominator of Equation 25.

FIG. 11C is a block diagram of a method and apparatus for implicitly calculating a stabilizing step size for the special case of F=I. In this case, Equation 24 becomes

$\begin{array}{cc}{\mu }^{\left[i\right]}=\frac{{\left(\underset{\_}{q}-{R\left({\Gamma }^{\left[i\right]}\right)}^2{\widehat{\underset{\_}{b}}}^{\left[i\right]}\right)}^H{\Gamma }^{\left[i\right]}\left(\underset{\_}{q}-R{\Gamma }^{\left[i\right]}{\widehat{\underset{\_}{b}}}^{\left[i\right]}\right)}{{\left(\underset{\_}{q}-R{\Gamma }^{\left[i\right]}{\widehat{\underset{\_}{b}}}^{\left[i\right]}\right)}^H{\left({\Gamma }^{\left[i\right]}\right)}^{H}R{\Gamma }^{\left[i\right]}\left(\underset{\_}{q}-R{\Gamma }^{\left[i\right]}{\widehat{\underset{\_}{b}}}^{\left[i\right]}\right)},& \mathrm{Equation}26\end{array}$

The signal β^[i] is generated by a Rake, maximal ratio combining, and despreading unit 1120 and multiplied 1121 by Γ^[i] to produce vector Γ^[i]β^[i]. A synthesis module 1122 processes the vector Γ^[i]β^[i] to produce a synthesized receive vector, which is norm-squared 1123 to produce the denominator of Equation 26.

A synthesized received signal is generated 1124 from the vector (Γ^[i])²β^[i] and processed with received signal y(t) by an adder 1125 to produce a difference signal. A Rake/combiner/despreader 1126 processes the difference signal to generate the vector q−R(Γ^[i])²{circumflex over (b)}^[i]. The inner product 1127 between this vector and the vector Γ^[i]β^[i] gives the numerator of Equation 26.

In an alternative embodiment, the stabilizing step size may be derived from the multipath channel gains,

$\begin{array}{cc}{\mu }^{\left[i\right]}=\mu =\max \left\{C,\left(\frac{\underset{\left(s\right),l}{\max }{{\alpha }_{\left(s\right),l}}^p}{\sum _{s=0}^{B-1}\sum _{l=0}^{L_{\left(s\right)}-1}{{\alpha }_{\left(s\right),l}}^p}\right)\right\},& \mathrm{Equation}27\end{array}$ where μ^[i] is fixed for every ICU and C, p and r are non-negative constants.

FIG. 12 is a block diagram of a symbol-estimation block comprising a plurality B of mixed-decision modules 1201-1203 configured to process signals received from B base stations. Mixed-decision module 1202 shows functionality that is common to all of the mixed-decision modules 1201-1203. De-biasing modules 1210-1211 scale each of a plurality K_(s) of input symbol estimates {tilde over (b)}_(s),k^[i+1] with a non-negative de-biasing constant d_(s),k^[i] for producing de-biased input symbol estimates. The mixed-decision module 1202 includes symbol-estimation modules 1212-1213 configured to perform symbol estimation on de-biased input symbol estimates whose constellations are known at the receiver.

The de-biasing constant may be expressed byd_(s),k^[i]=E[|b_(s),k|]/E└|{tilde over (b)}_(s),k^[i+1]|┘  Equation 28d_(s),k^[i]=√{square root over (E[|b_(s),k|²·]/E[|{tilde over (b)}{square root over (E[|b_(s),k|²·]/E[|{tilde over (b)}_(s),k^[i+1]|²])}  Equation 29d_(s),k^[i]=1 if the symbol constellation is unknown,  Equation 30where the statistical expectations may be approximated by time-averaging. De-biasing helps mitigate the “ping-pong” phenomenon often associated with iterative interference cancellation in which the symbol error rate oscillates with respect to iterations. After de-biasing, each value d_(s),k^[i]{tilde over (b)}_(s),k^[i+1] is operated on by a map Ψ_(s),k that takes the input into the complex plane to yield the updated symbol estimate{circumflex over (b)}_(s),k^[i+1]=Ψ_(s),k(d_(s),k^[i]{tilde over (b)}_(s),k^[i+1]).  Equation 31

The map Ψ_(s),k may be a mixed-decision map, which is a combination of soft and hard decisions. A soft-decision map is provided by a function Ψ_(s),k(x) that is a continuous function whose output ranges over the complex plane. Common examples, include, but are not limited to,

$\begin{array}{cc}{\Psi }_{\left(s\right),k}^{\mathrm{soft}}\left(x\right)=\{\begin{array}{c}{c}_{\left(s\right),k}x\\ \mathrm{or}\\ c_{\left(s\right),k}\left(\begin{array}{c}\tanh \left(a_{\left(s\right),k}\mathrm{Re}\left\{x\right\}\right)+\\ \sqrt{-1}\tanh \left(a_{\left(s\right),k}\mathrm{Im}\left\{x\right\}\right)\end{array}\right)\end{array}& \mathrm{Equation}32\end{array}$ for positive real-valued constants a_(s),k and c_(s),k. The expression Re{•} returns the real part of its argument, and Im{•} returns the imaginary part of its argument. A hard-decision map is provided when Ψ_(s),k(x) slices the input so that the output is an element from the complex symbol constellation employed by the k^th user of base station (s),Ψ_(s),k^hard(x)=slice(x).  Equation 33

The slicer quantizes its argument x to the nearest constellation symbol according to some metric (e.g., Euclidean distance). A hard decision is applicable only to those symbols whose constellations are known to the receiver.

A mixed-decision map Ψ_(s),k^mixed(x) produces an output that is a soft decision or a hard decision, such as

$\begin{array}{cc}{\Psi }_{\left(s\right),k}^{\mathrm{mixed}}\left(x\right)=\{\begin{array}{ccc}{\Psi }_{\left(s\right),k}^{\mathrm{hard}}\left(x\right)& \mathrm{if}& {\mathrm{SINR}}_{\left(s\right),k}>c_{\left(s\right),k}\\ {\Psi }_{\left(s\right),k}^{\mathrm{soft}}\left(x\right)& \mathrm{otherwise}& \end{array}& \mathrm{Equation}34\end{array}$

The mixed-decision map Ψ_(s),k^mixed(x) produces a hard decision if the SINR of a k^th user of base station (s) exceeds a threshold c_(s),k. Otherwise, a soft decision is performed. The SINR may be estimated with a time-averaged error-vector measurement (EVM). Time-averaging may cause a block of symbols to share the same SINR estimate.

An alternative mixed-decision map Ψ_(s),k^mixed(x) may act on individual symbols,

$\begin{array}{cc}{\Psi }_{\left(s\right),k}^{\mathrm{mixed}}(x)=\{\begin{array}{ccc}{\Psi }_{\left(s\right),k}^{\mathrm{hard}}\left(x\right)& \mathrm{if}& x\in C_{\left(s\right),k}\left(\mathrm{slice}\left(x\right)\right)\\ {\Psi }_{\left(s\right),k}^{\mathrm{soft}}\left(x\right)& \mathrm{otherwise}& \end{array},& \mathrm{Equation}35\end{array}$ where the constellation space for the symbol of a k^th user of base station (s) is partitioned into hard- and soft-decision regions with C_(s),k(b) denoting the hard-decision region for a symbol b from that user's constellation. If xεC_(s),k(b), then a hard decision for x is made. One embodiment for defining C_(s),k(b) is to include all points within a predetermined distance of b in the constellation space,C_(s),k(b)={x:distance(x,b)<c_(s),k(b)},  Equation 36where any distance metric may be used (e.g., |x−b|^p for some p>0) and the radii c_(s),k(b) over the set of constellation points b are chosen such that the hard-decision regions are non-overlapping. Alternative embodiments of the invention may employ different partitions of the constellation space. For example, edge constellation points may be given unbounded hard-decision regions.

Both the average SINR and instantaneous approaches are applicable to any known constellation; they need not be restricted to BPSK, QPSK, or even QAM. Either of these mixed-decision approaches may be performed with the additional constraint that the receiver knows only the constellation employed for a subset of the active codes. Such situations may arise in EV-DO and HSDPA networks. In such cases, the receiver may use soft decisions for codes employing an unknown modulation. Those skilled in the art will understand that a modulation classification of these codes may be performed, which may be particularly useful in systems wherein all interfering codes share the same unknown constellation.

The following algorithm, which is illustrated in FIG. 13, demonstrates one embodiment for performing IIC.

Algorithm 1:

• Purpose: Estimate the

$K=\sum _{s=0}^{B-1}{K}_{\left(s\right)}$ symbols in {circumflex over (b)}.

• Notation: The iteration index is represented by a superscript [i]; i=−1 is the initialization; i=0 corresponds to the output of the front-end processor; and
• i>0 corresponds to the i-th ICU.Definitions: b is in Equation 22
  • {circumflex over (b)}^[i] is in Equation 31
  • q is in Equation 17
  • R is in Equation 19
  • F is I or as in Equation 21
  • Γ^[i] is in Equation 14 with elements defined in Equation 3-Equation 5
  • μ^[i] is defined in Equation 23-Equation 27
  • Ψ maps each argument to a complex number to implement de-biasing as in Equation 28-Equation 30 and then symbol estimation as in Equation 32-Equation 36Initializations: {circumflex over (b)}^[−1]=0, a K×1 zero vector
  • Γ^[−1]=I, a K×K identity matrix
  • μ^[−1]=1Iterations: Index i=−1, 0, 1, . . . , M−1, where M is the number of times to iterate the succeeding update equationUpdate Equation: {circumflex over (b)}^[i+1]=Ψ{μ^[i](q−RΓ^[i]{circumflex over (b)}^[i])+FΓ^[i]{circumflex over (b)}^[i]}Output: {circumflex over (b)}={circumflex over (b)}^[M], the symbol estimates after M iterations of the update equation.

FIG. 13 shows an internal feedback loop comprising operations 1308, 1301, 1302, 1306, and an external feedback loop comprising operations 1308, 1301, 1303, and 1304. The output of the external feedback loop is q−RΓ^[i]{circumflex over (b)}^[i], which is multiplicatively scaled 1305 by μ^[i]. The scaled output is combined 1306 with the internal feedback loop to yield (q−RΓ^[i]{circumflex over (b)}^[i])+FΓ^[i]{circumflex over (b)}^[i], which is processed by a symbol estimator 1307 and fed to the iteration delay 1308 that begins the internal and external loops.

Although embodiments of the invention are described with respect to forward-link channels, embodiments may be configured to operate in reverse-link channels. In the reverse link, different users' transmissions experience different multipath channels, which requires appropriate modifications to Rake processing and signal synthesis. For example, a front-end processor may incorporate one Rake for every user in every base station rather than a single Rake per base station. Similarly, a separate multipath channel emulator may be employed for imparting multipath delays and gains to each user's signal. Accordingly, the number of constituent finger signals will equal the sum over the number of multipath fingers per user per base station, rather than the sum over the number of multipath fingers per base station.

It is clear that this algorithm may be realized in hardware or software and there are several modifications that can be made to the order of operations and structural flow of the processing.

Those skilled in the art should recognize that method and apparatus embodiments described herein may be implemented in a variety of ways, including implementations in hardware, software, firmware, or various combinations thereof. Examples of such hardware may include Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs), general-purpose processors, Digital Signal Processors (DSPs), and/or other circuitry. Software and/or firmware implementations of the invention may be implemented via any combination of programming languages, including Java, C, C++, Matlab™, Verilog, VHDL, and/or processor specific machine and assembly languages.

Computer programs (i.e., software and/or firmware) implementing the method of this invention may be distributed to users on a distribution medium such as a SIM card, a USB memory interface, or other computer-readable memory adapted for interfacing with a consumer wireless terminal. Similarly, computer programs may be distributed to users via wired or wireless network interfaces. From there, they will often be copied to a hard disk or a similar intermediate storage medium. When the programs are to be run, they may be loaded either from their distribution medium or their intermediate storage medium into the execution memory of a wireless terminal, configuring an onboard digital computer system (e.g., a microprocessor) to act in accordance with the method of this invention. All these operations are well known to those skilled in the art of computer systems.

The functions of the various elements shown in the drawings, including functional blocks labeled as “modules” may be provided through the use of dedicated hardware, as well as hardware capable of executing software in association with appropriate software. When provided by a processor, the functions may be performed by a single dedicated processor, by a shared processor, or by a plurality of individual processors, some of which may be shared. Moreover, explicit use of the term “processor” or “module” should not be construed to refer exclusively to hardware capable of executing software, and may implicitly include, without limitation, digital signal processor DSP hardware, read-only memory (ROM) for storing software, random access memory (RAM), and non-volatile storage. Other hardware, conventional and/or custom, may also be included. Similarly, the function of any component or device described herein may be carried out through the operation of program logic, through dedicated logic, through the interaction of program control and dedicated logic, or even manually, the particular technique being selectable by the implementer as more specifically understood from the context.

The method and system embodiments described herein merely illustrate particular embodiments of the invention. It should be appreciated that those skilled in the art will be able to devise various arrangements, which, although not explicitly described or shown herein, embody the principles of the invention and are included within its spirit and scope. Furthermore, all examples and conditional language recited herein are intended to be only for pedagogical purposes to aid the reader in understanding the principles of the invention. This disclosure and its associated references are to be construed as applying without limitation to such specifically recited examples and conditions. Moreover, all statements herein reciting principles, aspects, and embodiments of the invention, as well as specific examples thereof, are intended to encompass both structural and functional equivalents thereof. Additionally, it is intended that such equivalents include both currently known equivalents as well as equivalents developed in the future, i.e., any elements developed that perform the same function, regardless of structure.

Claims

1. An interference canceller configured for computing interference-cancelled symbol estimates, and performing processing during each of at least one iteration, the canceller comprising a decision module, the decision module configured for processing each of one or more input symbol estimates for producing one of a hard decision or a soft decision, and a de-biasing module configured for scaling the input symbol estimates with a scale factor to remove bias computed on the input symbol estimates for producing de-biased input symbol estimates.

2. The interference canceller recited in claim 1, wherein the de-biasing module is configured to set the scale factor equal to estimates of an expected ratio of symbol amplitudes to amplitudes of the input symbol estimates for each of the input symbol estimates having a known constellation.

3. The interference canceller recited in claim 2, wherein the de-biasing module is configured to produce de-biased symbol estimates characterized by

4. The interference canceller recited in claim 2, wherein the de-biasing module is configured to produce de-biased symbol estimates characterized by

5. The interference canceller recited in claim 1, wherein the de-biasing module is configured to set the de-bias scale factor of at least one symbol having a known constellation to unity.

6. An interference canceller configured for computing interference-cancelled symbol estimates, and performing processing during each of at least one iteration, the canceller comprising a decision module, the decision module configured for processing each of one or more input symbol estimates for producing one of a hard decision or a soft decision, wherein the decision module is configured for processing symbols with known constellations by partitioning a constellation space into hard- and soft-decision regions based on a function of proximity to nearby constellation points.

7. The interference canceller recited in claim 6, wherein the decision regions for at least one symbol are configured such that the hard-decision regions are collectively exhaustive so that only a hard decision is produced or the soft-decision regions are collectively exhaustive so that only a soft symbol decision is produced.

8. An interference cancellation method configured for computing interference-cancelled symbol estimates and providing for decisions, wherein providing for decisions comprises providing for processing each of one or more input symbol estimates for producing one of a hard decision or a soft decision, further comprising providing for de-biasing which is configured for de-biasing the input symbol estimates with a scale factor to remove bias computed on the input symbol estimates for producing de-biased input symbol estimates.

9. The method recited in claim 8, wherein providing for de-biasing is configured to set the scale factor equal to estimates of an expected ratio of symbol amplitudes to amplitudes of the input symbol estimates for each of the input symbol estimates having a known constellation.

10. The method recited in claim 9, wherein providing for de-biasing is configured to produce de-biased symbol estimates characterized by

11. The method recited in claim 9, wherein providing for de-biasing is configured to produce de-biased symbol estimates characterized by

12. The method recited in claim 8, wherein providing for de-biasing is configured to set the de-bias scale factor of at least one symbol having a known constellation to unity.

13. An interference cancellation method configured for computing interference-cancelled symbol estimates and providing for decisions, wherein providing for decisions comprises providing for processing each of one or more input symbol estimates for producing one of a hard decision or a soft decision, wherein providing for decisions is configured for processing symbols with known constellations by partitioning a constellation space into hard- and soft-decision regions based on a function of proximity to nearby constellation points.

14. The method recited in claim 13, wherein the decision regions for at least one symbol are configured such that the hard-decision regions are collectively exhaustive so that only a hard decision is produced or the soft-decision regions are collectively exhaustive so that only a soft symbol decision is produced.

15. An interference cancellation system configured for computing interference-cancelled symbol estimates and comprising a decision means operative during each of at least one iteration for each of one or more input symbol estimates, the decision means comprising a processing means configured for processing each of the one or more input symbol estimates for producing a hard decision or a soft decision, further comprising a de-biasing means configured for de-biasing the input symbol estimates with a scale factor to remove bias computed on the input symbol estimates for producing de-biased input symbol estimates.

16. The system recited in claim 15, wherein the de-biasing means is configured to set the scale factor equal to estimates of an expected ratio of symbol amplitudes to amplitudes of the input symbol estimates for each of the input symbol estimates having a known constellation.

17. The system recited in claim 16, wherein the de-biasing means is configured to produce de-biased symbol estimates characterized by

18. The system recited in claim 16, wherein the de-biasing means is configured to produce de-biased symbol estimates characterized by

19. The system recited in claim 15, wherein the de-biasing means is configured to set the de-bias scale factor of at least one symbol having a known constellation to unity.

20. The system recited in claim 15, wherein the decision means is configured for processing symbols with known constellations by partitioning a constellation space into hard- and soft-decision regions based on a function of proximity to nearby constellation points.

21. The system recited in claim 20, wherein the decision regions for at least one symbol are configured such that the hard-decision regions are collectively exhaustive so that only a hard decision is produced or the soft-decision regions are collectively exhaustive so that only a soft symbol decision is produced.

22. The interference canceller in claim 1, wherein the de-biasing module is configured to set the scale factor to unity for input symbol estimates having constellations that are unknown to the receiver.

23. The method recited in claim 8, wherein providing for de-biasing is configured to set the scale factor to unity for input symbol estimates having constellations that are unknown to the receiver.