Wireless receiver

Abstract

The present invention relates to a method and apparatus for demodulation in a wireless communications system transmitted across a wireless communications channel. The described wireless receiver includes a first antenna for receiving a wireless signal including a symbol transmitted across a wireless communications channel perceived by the first antenna, an observation modifier for generating a modified observation (y) of the symbol based on a product of the received observation (r) and the complex conjugate of a channel estimate (h*), a log-likelihood ratio (LLR) module generating log-likelihood ratios (LLRs) based on the modified observation and the channel estimate, and a maximum-likelihood-based decoder for decoding the symbol based on the LLRs.

Claims

1. A method comprising: receiving, at a first antenna, an observation of a symbol transmitted across a wireless communications channel perceived by the first antenna; generating a modified observation of the symbol based on a product of the received observation and the complex conjugate of a channel estimate of the channel; and generating, based on the modified observation and the channel estimate, log-likelihood ratios (LLRs) for the symbol for a maximum-likelihood-based decoder to decode, wherein generating log-likelihood ratios includes: generating a LLR associated with a most significant bit of the symbol, and generating a LLR associated with a next most significant bit of the symbol; wherein the log-likelihood ratio (LLR) associated with the most significant bit is generated based on the ratio of the real part (y.sub.I) or imaginary part (y.sub.Q) of the modified observation to Gaussian-distributed noise power of the channel.

2. The method of claim 1 wherein the LLR associated with the most significant bit is estimated to be 2y.sub.I/.sup.2 or 2y.sub.Q/.sup.2, where .sup.2 represents the Gaussian-distributed noise power of the channel.

3. The method of claim 1 wherein the LLR associated with the most significant bit is estimated to be 4y.sub.I/.sup.2 or 4y.sub.Q/.sup.2, where .sup.2 represents the Gaussian-distributed noise power of the channel.

4. The method of claim 1 wherein the LLR associated with the most significant bit is estimated to be 4y.sub.I/.sup.24|h|.sup.2/.sup.2 or 4y.sub.Q/.sup.24|h|.sup.2/.sup.2, where .sup.2 represents the Gaussian-distributed noise power of the channel and |h| represents the magnitude of the channel estimate.

5. The method of claim 1 wherein the LLR associated with the most significant bit is 4y.sub.I/.sup.2+4|h|.sup.2/.sup.2, where .sup.2 represents the variance of the Gaussian-distributed noise power of the channel and |h| represents the magnitude of the channel estimate.

6. The method of claim 1 wherein the LLR associated with the next most significant bit is generated based on the ratio of the real part (y.sub.I) or imaginary part (y.sub.Q) of the modified observation to Gaussian-distributed noise power of the channel.

7. The method of claim 6 wherein the LLR associated with the most significant bit is 4|h|/.sup.22y.sub.I/.sup.2 or 4|h|/.sup.22y.sub.Q/.sup.2, where .sup.2 represents the variance of the Gaussian-distributed noise power of the channel and |h| represents the magnitude of the channel estimate.

8. The method of claim 1 wherein the Gaussian-distributed noise power is estimated or measured.

9. The method of claim 1 further comprising receiving, at a second antenna located separately from the first antenna, another observation of the symbol transmitted across a wireless communications channel perceived by the second antenna, and wherein generating a modified observation includes generating the modified observation also based on a product of the other received observation and the complex conjugate of a channel estimate of the channel perceived by the second antenna.

10. The method of claim 9 wherein generating the modified observation is based on a weighted sum of (a) the product of the received observation and the complex conjugate of a channel estimate of the channel and (b) the product of the other received observation and the complex conjugate of a channel estimate of the channel perceived by the second antenna.

11. The method of claim 10 wherein the weighted sum is based on weights associated with the noise power as observed by the respective antennas.

12. The method of claim 1 further comprising decoding, by a maximum-likelihood-based decoder, the LLRs.

13. A wireless receiver comprising: a first antenna for receiving an observation including a symbol transmitted across a wireless communications channel perceived by the first antenna; an observation modifier for generating a modified observation of the symbol based on a product of the received observation and the complex conjugate of a channel estimate of the channel; a log-likelihood ratio (LLR) for the symbol module generating log-likelihood ratios (LLRs) based on the modified observation, wherein generating log-likelihood ratios includes: generating a LLR associated with a most significant bit of the symbol, and generating a LLR associated with a next most significant bit of the symbol; wherein the log-likelihood ratio (LLR) associated with the most significant bit is generated based on the ratio of the real part (y.sub.I) or imaginary part (y.sub.Q) of the modified observation to Gaussian-distributed noise power of the channel; and a maximum-likelihood-based decoder for decoding the symbol based on the LLRs.

14. The wireless receiver of claim 13 further comprising a second antenna for receiving another observation of the symbol transmitted across a wireless communications channel by the second antenna, the first antenna and the second antenna are located to provide spatial diversity, wherein the observation modifier is configured to generate the modified observation based on weighted sum of (a) the product of the received observation and the complex conjugate of a channel estimate of the channel and (b) the product of the other received observation and the complex conjugate of a channel estimate of the channel perceived by the second antenna.

15. The wireless receiver of claim 14 wherein the weighted sum is based on weights associated with the noise power as observed by the respective antennas.

16. A non-transitory machine-readable medium comprising instructions executable by one or more processors, the instructions including the steps of: receiving, at a first antenna, an observation of a symbol transmitted across a wireless communications channel perceived by the first antenna; generating a modified observation of the symbol based on a product of the received observation and the complex conjugate of a channel estimate of the channel; and generating, based on the modified observation and the channel estimate, log-likelihood ratios (LLRs) for the symbol for a maximum-likelihood-based decoder to decode, wherein generating log-likelihood ratios includes: generating a LLR associated with a most significant bit of the symbol, and generating a LLR associated with a next most significant bit of the symbol; wherein the log-likelihood ratio (LLR) associated with the most significant bit is generated based on the ratio of the real part (y.sub.I) or imaginary part (y.sub.Q) of the modified observation to Gaussian-distributed noise power of the channel.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) FIG. 1 illustrates a schematic diagram of a wireless communications system including a transmitter and a receiver.

(2) FIGS. 2A and 2B illustrate simulated 16-QAM constellation diagrams of (A) equalised received symbols and (B) LLR-processed (log-likelihood-ratio-processed) received symbols under high channel-power-to-noise-power, or signal-to-noise, ratio (SNR).

(3) FIGS. 3A and 3B illustrate simulated 16-QAM constellation diagrams of (A) equalised received symbols and (B) LLR-processed received symbols under medium channel-power-to-noise-power, or signal-to-noise, ratio (SNR).

(4) FIGS. 4A and 4B illustrate simulated 16-QAM constellation diagrams of (A) equalised received symbols and (B) LLR-processed received symbols under low channel-power-to-noise-power, or signal-to-noise, ratio (SNR).

(5) FIG. 5 illustrates bit to constellation mapping of 16-QAM system.

(6) FIG. 6A illustrates a wireless transmitter and of a wireless receiver coupled by a communication channel in accordance with one embodiment.

(7) FIG. 6B illustrates a logic block diagram of an example of a subsystem 600 in accordance with one embodiment.

(8) FIG. 6C illustrates a method of demodulation in accordance with one embodiment.

(9) FIG. 6D illustrates a wireless transmitter and of a wireless receiver coupled by a communication channel in accordance with another embodiment.

(10) FIG. 7A plots a comparison between optimal LLR values and approximated LLR values for b0 for small-magnitude y.sub.I.

(11) FIG. 7B plots a comparison between optimal LLR values and approximated LLR values for b0 for large-magnitude y.sub.I.

(12) FIG. 7C plots a comparison between optimal LLR values and approximated LLR values for b1.

(13) FIG. 7D plots a comparison between optimal LLR values and folding LLR values for b0.

(14) FIG. 7E plots the comparison shown in FIGS. 7A, 7B, 7C and 7D together.

(15) FIG. 8 plots a comparison of the packet error rate between using small-magnitude approximation only and using both small and large-magnitude approximations to LLR.

(16) FIG. 9 illustrates a logic block diagram of an example of a demodulator in accordance with one embodiment.

DETAILED DESCRIPTION OF EMBODIMENTS

(17) Equalisation in Varying Channels

(18) The inventors have recognised that the equalisation characterised by Equation (2) may give rise to degradation in decoding performance if the channel is time-varying due to, for example, changing multipath effects arising from movements of the receiver, the transmitter and/or any surrounding objects.

(19) Referring to Equation (1), the symbol d may be formed at the transmitter based on a sequence of coded bits. For example, 1, 2, 4 or 6 coded bits may map to binary phase-shift keying (BPSK), quadrature phase-shift keying (QPSK), 16-QAM (quadrature amplitude modulation) or 64-QAM constellations, respectively. Noise n is characterised by its power, and may be estimated or measured at the receiver. For receivers implementing maximum-likelihood based decoding, such as Viterbi decoding, the inventors recognise that degradation in decoding performance can arise if the noise statistics of the inputs to the Viterbi decoder are not constant over the received coded bits. The degradation can be attributed to the reliance of maximum-likelihood based estimation of the bit sequence on the product of Gaussian-distributed noise statistics. An illustrative example using BPSK is described as follows.

(20) In the case of BPSK, the Gaussian-distributed probability of the additive white Gaussian noise (AWGN) for each of the two possible symbols for a respective subcarrier j can be characterised by:

(21) $\begin{array}{cc}C\exp \left[\frac{{.Math.r_i-h_i{d}_i.Math.}^2}{2{}^2}\right]& \left(3\right)\end{array}$

(22) over all bits constituting the bit sequence, where i now indexes the bit position in the sequence, C is a constant, .sup.2 represents the noise power and |r.sub.ih.sub.id.sub.i| represents the noise-induced deviation in the I-Q (in-phase and quadrature) plane between the received bit (r.sub.i) and the transmitted bit affected by a noise-free channel (h.sub.id.sub.i). For higher order constellations, a similar sequence metric can be formed.

(23) In practice, Viterbi decoding is executed in the logarithmic domain. Instead of computing a product of exponentials, therefore, a sum of exponents is computed to provide the Viterbi metric in the following form:

(24) $\begin{array}{cc}{.Math.}_i\left[\frac{{.Math.r_i-h_i{d}_i.Math.}^2}{2{}^2}\right]& \left(4\right)\end{array}$

(25) where, for the purposes of calculating the metric, C is ignore since it is a constant for all sequence {d.sub.i} considered. Viterbi decoding involves, among other steps, considering all possible bit sequences and selecting the one with the highest likelihood (or least distance in the logarithmic domain). If is constant, it too can be disregarded as it will not affect the selection based on sequence metrics. In practice, the Viterbi decoding may therefore be reduced to accumulators of the inputs, such that they can be run faster to inspect the possible bit sequences, albeit in a highly structured manner.

(26) If a received observation is equalised, the equalised observation y according to Equation (2) is given by
y.sub.equalised=r/h=d+n/h(5)

(27) In the case where the channel h is varying in amplitude (e.g. in the case of fading channels), the effective noise power presented at the input of the Viterbi decoder (i.e. .sup.2 scaled based on h) is no longer identically distributed from bit to bit and the Viterbi metric cannot be simplified as above without loss of performance. To illustrate such a loss in performance, FIGS. 2A-4B illustrate simulated 16-QAM constellation diagrams of (a) received symbols equalised according to Equation (2) (labelled Equalised) and (b) received symbols processed as log-likelihood ratios (LLR) in preparation for maximum-likelihood based decoding (labelled Maximum-likelihood and using a small-amplitude LLR approximation for the most significant bits described later) under high, medium, and low channel-power-to-noise-power, or signal-to-noise, ratio (SNR) at the receiver. Each dot plotted in the I-Q plane represents one of multiple received symbols constructed from a random sequence of input bits transmitted over a fading channel. The scattered pattern at each of the 16 constellation points (see, for example, FIG. 2A) arises from the AWGN. The vertical and horizontal scales of these plots are arbitrary. The received symbols are assumed to be transmitted via a fading channel. As illustrated in FIGS. 2A and 2B, under high SNR (around 20 dB), both (a) and (b) have a recongisable 16-QAM constellation structure. Under medium SNR (around 10 dB), only (a) has a recongisable 16-QAM constellation structure, as seen in FIGS. 3A and 3B. Under low SNR (around 5 dB), neither (a) nor (b) has a recongisable 16-QAM constellation structure, as seen in FIGS. 4A and 4B.

(28) Having recognised the effects of varying effective noise from bit to bit on maximum-likelihood based decoding performance, the inventors have devised a method of demodulation based on a channel-matched observation to address these effects. The present disclosure is generally applicable to any QAM schemes, although each QAM scheme may be more efficiently implementation due to specific bit-to-constellation mapping. The following description discloses bit-to-constellation mapping using 16-QAM as an example. Similar bit-to-constellation mapping may be derived for higher-order QAM schemes.

(29) Bit to Constellation Mapping

(30) In 16-QAM, constellation symbols d are defined by four bits in the ordered sequence of b0, b1, b2 and b3, each representing a logical value of either 0 or 1, as shown in FIG. 5, and taking a real value of either 1 or 1, respectively, as used in Equations (6a)-(8b) below.

(31) The real part d.sub.I (or the I-component) and the imaginary part d.sub.Q (or Q-component) of the symbol d may be given by:
d.sub.I=2b0b0b1=b0(2b1)(6a)
d.sub.Q=2b2b2b3=b2(2b3)(6b)

(32) The absolute values of each component are given by:
|d.sub.I|=|b0(2b1)|=|2b1|=2b1(7a)
|d.sub.Q|=|b2(2b3)|=|2b3|=2b3(7b)

(33) Re-arranging (7a) and (7b) provides:
b1=2|d.sub.I|(8a)
b3=2|d.sub.Q|(8b)

(34) Equations (8a) and (8b) therefore illustrate that it is possible to recover the least significant bits b1 and b3 from the absolute value of the I and Q components.

(35) In 16-QAM, b0 and b2 are referred to as the most significant bits, whereas b1 and b3 are referred to as the next (or the least) significant bits. In 64-QAM (where a symbol constitutes 6 bits in the ordered sequence of b0, b1, b2, b3, b4 and b5), b0 and b3 are the most significant bits, b1 and b4 are the next significant bits and b2 and b5 are the least significant bits.

(36) A Wireless Receiver and a Demodulator

(37) Disclosed herein is a wireless receiver 220 as illustrated in FIG. 6A. The described wireless receiver 220 includes a first antenna 221 for receiving a wireless signal including a symbol transmitted across a wireless communications channel 250 perceived by the first antenna, an observation modifier 222 for generating a modified observation (v) of the symbol based on a product of the received observation (r) and the complex conjugate of a channel estimate (h*), a log-likelihood ratio (LLR) module 225 generating log-likelihood ratios (LLRs) based on the modified observation and the channel estimate, and a maximum-likelihood-based decoder 226 for decoding the symbol based on the LLRs. Together, the observation modifier 222 and the LLR module 225 may be referred to as a demodulator 223. The receiver 220 may be configured to perform a demodulation method 300 as illustrated in FIG. 6C. The outputs of the method 300 are log-likelihood ratios (LLRs) based on modified observations of a symbol for subsequent decoding by a maximum-likelihood-based decoder, such as 226. The described method 300 includes the step 302 of receiving, at a first antenna 221, an observation of a symbol transmitted across a wireless communications channel 250, the step 304 of generating a modified observation based on a product of the received observation (r) and the complex conjugate of a channel estimate (h*), and the step 306 of generating, based on the modified observation, log-likelihood ratios (LLRs) for a maximum-likelihood-based decoder, such as the decoder 226, to decode. Steps 304 and 306 may be carried out by the demodulator 223.

(38) The wireless communications channel 250 is one that is perceived by the first antenna 221. In some arrangement, as illustrated in FIG. 6D, wireless receiver 220 includes a second antenna 221 located in relation to the first antenna 221 to provide spatial diversity. The second antenna 221 when so placed perceives a wireless communications channel 250 that is different from the channel 250 perceived by the first antenna and is associated with a different channel estimate 230. In general, the first antenna 221 and the second antenna 221 are located at least approximately a wavelength of the wireless signal apart for spatial diversity.

(39) In one arrangement, the wireless receiver 220 further includes, for each antenna, a channel estimator 230 for providing the channel estimate and a noise power estimator 240 for estimating or measuring the noise power of the perceived channel. As described further below, the observation modifier 222 uses the channel estimate to generate the modified observation, whereas the LLR module 225 uses the channel estimate and the estimated or measured noise power to generate LLRs. A wireless receiver focussing on a single-antenna implementation is first described. A wireless receiver generalising to a multiple-antenna (i.e. spatially diversified) implementation is then described.

(40) Channel-Matching an Observation

(41) In the step 304, the observation modifier 222 is configured to generate a modified observation y. The observation modifier 222 computes the modified observation y based on a product of the received observation r and the complex conjugate of the complex-valued channel estimate h*. In one implementation, the modified observation y is given by:
y=rh*(9)

(42) In another implementation, y is given by=crh*, that is, multiplied by an additional constant c compared to the implementation given in Equation (9). In the description that follows, the modified observation y is assume to be given by Equation (9).

(43) As will become apparent below, the modified observation y is channel-matched since the log-likelihood ratios calculated based on the modified observation y have reduced dependence on the amplitude of the complex-valued channel h, compared to the equalised observation given by Equation (2). Stronger channels should generally be more assertive than weaker channels. The observation is brought into coherent alignment with the original transmitted symbol, where the rotational effects because of the channels are removed.

(44) Generating Log Likelihood Ratios

(45) In step 306, the modified (e.g. channel-matched) observation generated may be provided to the LLR module 225 for generating LLRs. The generated LLRs may be used for subsequent decoding by, for example, decoder 226. In one arrangement, the LLR module 225 computes the LLR for individual bits constituting the symbol d based on the modified observation y. Using the case of 16-QAM as an example, the probability P that the received observation r resulted from transmission with the i-th bit b.sub.i (where i0, 1, 2 and 3) defining the transmitted symbol d and being 1 is:
P(r|b.sub.i=1)=.sub.dD.sub.i.sub.1P(r|d)(10)

(46) where D.sub.i.sup.1 is the set of all constellation points with bit i (b.sub.i) equal to 1. Expanding and simplifying (10) provides:

(47) $\begin{array}{cc}P\left(r.Math.b_i=1\right)=C\underset{d{D}_i^1}{.Math.}\exp \left[\frac{\mathrm{Re}\left(y^{*}d\right)-\frac{{.Math.h.Math.}^2{.Math.d.Math.}^2}{2}}{{}^2}\right]& \left(11\right)\end{array}$

(48) where y=rh* is the matching of the observation to the channel. A similar expression to Equation (11) can be derived where the additional constant c is involved.

(49) Decomposing the symbol constellation point d and the modified observation y into their in-phase (I) and quadrature (Q) components as d=d.sub.I+jd.sub.Q and y=y.sub.I+jy.sub.Q, respectively, Equation (11) becomes:

(50) $\begin{array}{cc}P\left(r|b_i=1\right)=C\underset{d_I{M}_i^1}{\overset{}{.Math.}}\exp \left[\frac{y_I{d}_I-\frac{{.Math.h.Math.}^2{.Math.d_I.Math.}^2}{2}}{{}^2}\right]\underset{d_Q}{\overset{}{.Math.}}\exp \left[\frac{y_q{d}_q-\frac{{.Math.h.Math.}^2{.Math.d_Q.Math.}^2}{2}}{{}^2}\right]& \left(12\right)\end{array}$

(51) where the bit to constellation mapping according to (8a) and (8b) have been used. The set of real value M.sub.i.sup.1 are the values of each axis (assumed the same for I and Q components) where bit i (b.sub.i) takes value 1. For example, in 16-QAM. M.sub.1.sup.1=M.sub.3.sup.1={1, +1} and M.sub.0.sup.1=M.sub.2.sup.1={+1, +3}. It is assumed here that the bit of interest affects the in-phase component. The sum over d.sub.Q is the sum over all possible values of the quadrature component. A similar expression to Equation (12) can be derived for the probability P(r|b.sub.i=0) that r resulted from transmission with the i-th bit b.sub.i (where i0, 1, 2 and 3) defining the transmitted symbol d and being 0.

(52) The likelihood ratio, assuming that bit i affects only the in-phase components, is given by:

(53) $\begin{array}{cc}\frac{P\left(r|b_i=1\right)}{P\left(r|b_i=0\right)}=\frac{\underset{d_I{M}_i^1}{\overset{}{.Math.}}\exp \left[\left(y_I{d}_I-\frac{{.Math.h.Math.}^2{.Math.d_I.Math.}^2}{2}\right)/{}^2\right]}{\underset{d_I{M}_i^0}{\overset{}{.Math.}}\exp \left[\left(y_I{d}_I-\frac{{.Math.h.Math.}^2{.Math.d_I.Math.}^2}{2}\right)/{}^2\right]}& \left(13\right)\end{array}$

(54) b.sub.0 For the most significant bit (MSB) b0 of the 16-QAM constellation, which relates to the sign of the transmitted symbol, M.sub.0.sup.1={+1, +3} and M.sub.0.sup.0={1, 3}, which provides the likelihood ratio for bit b0 as:

(55) $\begin{array}{cc}\frac{P\left(r|b_0=1\right)}{P\left(r|b_0=0\right)}=\frac{\exp \left[\left(y_I-\frac{{.Math.h.Math.}^2}{2}\right)/{}^2\right]+\exp \left[\left(3y_I-\frac{9{.Math.h.Math.}^2}{2}\right)/{}^2\right]}{\exp \left[\left(-y_I-\frac{{.Math.h.Math.}^2}{2}\right)/{}^2\right]+\exp \left[\left(-3y_I-\frac{9{.Math.h.Math.}^2}{2}\right)/{}^2\right]}& \left(14\right)\end{array}$

(56) where the first and second exponents in the numerator arise from the symbol constellations d.sub.I having real values M=+1 and M=+3, respectively, and the first and second exponents in the denominator arise from the symbol constellations d.sub.I having real values M=1 and M=3, respectively.

(57) Taking the logarithmic of Equation (14), the log likelihood ratio .sub.0 for b0 is given by:

(58) $\begin{array}{cc}{}_0=\log \frac{P\left(r|b_0=1\right)}{P\left(r|b_0=0\right)}& \left(15\right)\end{array}$

(59) Equation (15) can be simplified by approximation to quicken execution time of the demodulator 223. There are four exponents in Equation (14). Depending on the value of the modified observation y (in particular its 1-component y.sub.I or its Q-component y.sub.Q), only the exponent having greater contribution to the probability function in each of the numerator and the denominator is retained, while the exponent having less contribution to the probability function in each of the numerator and the denominator is disregarded. In particular: For small-magnitude modified observations, that is for y.sub.I<2, .sub.0 can be approximated by disregarding the contributions from the M=+3 and M=3 symbol constellations (i.e. the second exponents in both the numerator and the denominator in (14)) and relying on the contributions from M=+1 and M=1 (i.e. the second exponents in both the numerator and the denominator in (14)) to give:

(60) $\begin{array}{cc}{}_0\left[\frac{\left(y_I-\frac{{.Math.h.Math.}^2}{2}\right)}{{}^2}\right]-\left[\frac{\left(-y_I-\frac{{.Math.h.Math.}^2}{2}\right)}{{}^2}\right]=2y_I/{}^2& \left(16a\right)\end{array}$ For positive and large-magnitude modified observations, that is for y.sub.I>+2, .sub.0 can be approximated by disregarding the contributions from M=+1 and M=3 (i.e. the second and the first exponents in the numerator and the denominator, respectively in (14)) and relying on the contributions from M=1 and M=+3 (i.e. the first and the second exponents in the numerator and the denominator, respectively, in (14)) to give:

(61) $\begin{array}{cc}{}_0\left[\frac{\left(y_I-\frac{{.Math.h.Math.}^2}{2}\right)}{{}^2}\right]-\left[\frac{\left(-3y_I-\frac{9{.Math.h.Math.}^2}{2}\right)}{{}^2}\right]=\frac{4y_I}{{}^2}-\frac{4{.Math.h.Math.}^2}{{}^2}& \left(16b\right)\end{array}$ For negative and large-magnitude modified observations, that is for y.sub.I<2, .sub.0 can be approximated by disregarding the contributions from M=1 and M=+3 (i.e. the first and the second exponents in the numerator and the denominator, respectively, in (14)) and relying on the contributions from M=+1 and M=3 (i.e. the second and the first exponents in the numerator and the denominator, respectively, in (14))) to give:

(62) 0$\begin{array}{cc}{}_0-\left[\frac{\left(-y_I-\frac{{.Math.h.Math.}^2}{2}\right)}{{}^2}\right]-\left[\frac{\left(-3y_I-\frac{9{.Math.h.Math.}^2}{2}\right)}{{}^2}\right]=\frac{4y_I}{{}^2}+\frac{4{.Math.h.Math.}^2}{{}^2}& \left(16c\right)\end{array}$ For completeness, for equal power modified observations, .sub.04y.sub.I/.sup.2 by disregarding the contribution of M=+/1 and relying on the contribution of M=+/3.

(63) FIGS. 7A and 7B provide a comparison between the optimal LLR (i.e. the LLR given by Equation (14)) and the approximations given by Equations (16a)-(16c). In FIG. 7A, the optimal LLR0 (.sub.0) (reference numeral 702) is seen to be closely approximated by the approximated LLR0 (.sub.0) given by Equation (16a) (reference numeral 704) at small magnitudes (i.e. y.sub.I2). Similarly, in FIG. 7B, the optimal LLR0 (.sub.0) (reference numeral 702) is seen to be also closely approximated by the approximated LLR0 (.sub.0) given by Equations (16b) and (16c) (reference numerals 706a and 706b) at large magnitudes (i.e. y.sub.I>2). (Note that the approximated LLR0 (.sub.0) at y.sub.I=0 is 4|h|.sup.2/.sup.2 and is thus discontinuous. In FIG. 7B, the small segment 706c joining the two segments 706a and 706b is an artefact resulting from the graphing tool)

(64) A generated LLR greater than 0 (i,e. above the upper part of Figure) indicates that the bit b.sub.0 in question is more likely to be a logical 1 than a logical 0 (and vice versa). The LLRs are provided to the decoder 226 for decoding purposes. The close agreement between the optimal and the approximated LLRs indicates, in some arrangements, that it may be beneficial to sacrifice obtaining the optimal values of LLR for the sake of faster execution of LLR computation according to Equations (16a) to (16c). For completeness, for y.sub.I equal to boundary values (e.g. y.sub.I=2), since the small-magnitude and the large-magnitude approximations (for small |h|) intersect at y.sub.I=2, as shown in FIG. 7A, either approximation is appropriate.

(65) b1

(66) For the next significant bit b1 of the 16-QAM constellation. M.sub.1.sup.1={1, +1} and M.sub.1.sup.0={3, +3}, which provides the likelihood ratio for bit b1 as:

(67) $\begin{array}{cc}\frac{P\left(r|b_1=1\right)}{P\left(r|b_1=0\right)}=\exp \left[4\frac{{.Math.h.Math.}^2}{{}^2}\right]\frac{\exp \left[-y_I/{}^2\right]+\exp \left[y_I/{}^2\right]}{\exp \left[-3y_I/{}^2\right]+\exp \left[3y_I/{}^2\right]}& \left(17\right)\end{array}$

(68) where the first and second exponents in the numerator arise from the symbol constellations d.sub.I having real values M=1 and M=+1, respectively, and the first and second exponents in the denominator arise from the symbol constellations d.sub.I having real values M=3 and M=+3, respectively.

(69) Taking the logarithmic of Equation (17), the log likelihood ratio .sub.1 for b1 is given by:

(70) $\begin{array}{cc}{}_1=\log \frac{P\left(r|b_1=1\right)}{P\left(r|b_1=0\right)}& \left(18\right)\end{array}$

(71) Like Equation (15), Equation (18) can be simplified by approximation in a similar manner to quicken execution time as follows: For positive modified observations, that is for y.sub.I>0, .sub.1 can be approximated by disregarding the contributions from the M=1 and M=3 symbol constellations (i.e. the first exponents in both the numerator and the denominator in (17)) and relying on the contributions from M=+1 and M=+3 (i.e. the second exponents in both the numerator and the denominator in (17)) to give:

(72) $\begin{array}{cc}{}_1\left[\frac{4{.Math.h.Math.}^2}{{}^2}\right]+\left[y_I/{}^2\right]-\left[3y_I/{}^2\right]=\left[\frac{4{.Math.h.Math.}^2}{{}^2}\right]-\frac{2y_I}{{}^2}& \left(19a\right)\end{array}$ For negative modified observations, that is for y.sub.I<0, .sub.1 can be approximated by disregarding the contributions from M=+1 and M=+3 (i.e. the second exponents in both the numerator and the denominator in (17)) to give and relying on the contributions from M=1 and M=3 (i.e. the first exponents in both the numerator and the denominator in (17)):

(73) $\begin{array}{cc}{}_1\left[\frac{4{.Math.h.Math.}^2}{{}^2}\right]+\left[-y_I/{}^2\right]-\left[-3y_I/{}^2\right]=\left[\frac{4{.Math.h.Math.}^2}{{}^2}\right]+\frac{2y_I}{{}^2}& \left(19b\right)\end{array}$ Combining Equations (19a) and (19b), the LLR for b1 can be approximated and expressed as:

(74) $\begin{array}{cc}{}_1\left[\frac{4{.Math.h.Math.}^2}{{}^2}\right]-.Math.\frac{2y_I}{{}^2}.Math.& \left(19c\right)\end{array}$

(75) .sub.1(y.sub.I=0)=.sub.0.sup.large-magnitude(y.sub.I=0)=4|h|.sup.2/.sup.2=4 is termed the fold point of the demodulator 223, which is called a folding demodulator. That is, the fold point takes the value of the LLR of the next significant bit .sub.1 (or the negative of the most significant bit in large-magnitude approximation .sub.0.sup.large-magnitude) when the in-phase component or quadrature component (depending on which of complementary LLRs is in question) of the modified observation is zero. The fold point may assist in identifying the LLR for the next significant bit. =|h|.sup.2/.sup.2 is the channel power to noise power ratio, or the signal-to-noise ratio (SNR). The noise power, represented by .sup.2, may be estimated or measured by the noise power estimator 240.

(76) FIG. 7C provides a comparison between the optimal LLR (i.e. Equation (17)) and the approximations given by Equations (19a)-(19c). The optimal LLR1 (.sub.1) (reference numeral 708) is seen to be closely approximated by the approximated LLR1 (.sub.1) given by Equation (19c) (reference numeral 710) regardless of magnitudes of y.sub.I.

(77) A LLR greater than 0 (i.e. the upper part of FIG. 7C) indicates that the corresponding bit b.sub.1 is more likely to be a logical 1 than a logical 0 (and vice versa). The LLRs are provided to the decoder 226 for decoding purposes. The close agreement indicates, in some arrangements, that it may be beneficial to sacrifice obtaining the optimal values of LLR for faster execution of LLR computation according to Equations (19a) to (19c).

(78) For completeness, FIG. 7D provides a comparison between the optimal LLR0 (.sub.0) (reference numeral 702) and the folding LLR0 (i.e. equal power modified observation) .sub.0=4y.sub.I/.sup.2 (reference numeral 712). For illustration purposes, FIG. 7E provides all LLR0 shown in FIGS. 7A-7D on the same scale. In all of FIGS. 7A-7E, the signal-to-noise ratio is 0 dB, such that =|h|.sup.2/.sup.2=1.

(79) b2 and b3

(80) Similarly, a skilled person in the art would appreciate that, the expressions .sub.2 and .sub.3 for the LLRs for bit b2 and b3 are similar to those of .sub.0 and .sub.1, respectively, except that y.sub.I is replaced by y.sub.Q. .sub.2 and .sub.0 are complementary LLRs, and are the most significant bits. Similarly, .sub.3 and .sub.1 are also complementary LLRs, and are the next significant bits (and also the least significant bits in the case of 16-QAM)).

(81) FIG. 6B illustrates a logic block diagram of a subsystem 600 including an observation modifier 222 and related components in facilitating calculation of Equations (16a)-(16c) and Equations (19a)-(19c). In the first branch 602, the observation modifier 222 obtains a received observation r. The observation modifier 222 multiplies the received observation r by the conjugate of the channel estimate h obtained from the second branch 604 from, for example, a channel estimator (not shown) to generate a modified observation y. Either the I or Q component of y, depending on which of the complementary LLRs is to be calculated, is saturated by a saturate block 608. In the second branch 604, the channel estimate is conjugated by a conjugate block 610 and provided to the first branch 602 for the multiplication operation described above. Still in the second branch 604, the channel estimate is multiplied by its conjugate before being saturated by a saturate block 612 to provide the channel power (represented by |h|.sup.2). In the third branch 606, a measured or estimated noise power (represented by .sup.2) is obtained, for example from a noise power estimator, and is inverted by an inverse block 618. The inverse block 618 may be a look up table (LUT). The inverted noise power (1/.sup.2) is provided to the second branch 604 for multiplying with the channel power to estimate the signal-to-noise ratio (represented by |h|.sup.2/.sup.2). The inverted noise power (1/.sup.2) is provided to the first branch 602 to calculate y.sub.I/.sup.2 or y.sub.Q/.sup.2. The outputs of the first branch 602 (i.e. y.sub.I/.sup.2 or y.sub.Q/.sup.2 and the second branch (i.e. |h|.sup.2/.sup.2) facilitate the calculation of LLR approximations given by Equations (16a)-(16c) and also (19a)-(19c).

(82) From the foregoing, it is apparent that the LLRs generated based on the modified observation v has reduced dependence on the magnitude of the channel h.

(83) 64 QAM

(84) The foregoing examples are directed to a 16-QAM system. It should be apparent to a skilled person to derive, for a 64-QAM systems, similar expressions for .sub.0, .sub.1, .sub.2, .sub.3, .sub.4 .sub.5.

(85) Further Simplification

(86) Equation (19c) indicates that .sub.1 (and hence .sub.3) is given by one approximation. On the other hand, Equations (16a)-(16c) indicate that .sub.0 (and hence .sub.2) can be given by three different approximations, depending on the magnitude of the real part (and the imaginary part) of the modified observation y. It would further simplify the demodulator 223 where .sub.0 (and hence .sub.2) is approximated by only one of the approximations. FIG. 8 illustrates a comparison between the packet error rate (PER) 802 where .sub.0 (and hence .sub.2) is approximated by all 3 approximations depending on the magnitude of the real part (or the imaginary part) of the modified observation, and the PER 804 where .sub.0 (and hence .sub.2) is approximated by Equation (16a), that is, the small-magnitude approximation, regardless of on the magnitude of the real part (or the imaginary part) of the modified observation y. .sub.1 and .sub.3 are approximated by Equation (19c). As illustrated in FIG. 8, there is only a small degradation in PER where only the small-magnitude approximation of .sub.0 and .sub.2 is used. For example, at receiver power of 89 dBm, the difference is between about 0.014 (for PER 802) and 0.015 (for PER 804).

(87) Spatial Diversity

(88) For spatially diversified receivers having multiple antennas, such as a first antenna 221 and a second antenna 221, each antenna generally perceives a different channel h and is associated with a different antenna noise power n. The equations above may be extended with modifications as follows and the demodulation may be adapted accordingly. For the a-th antenna, the received observation r of symbol d becomes:
r[i,j,a]=h[i,j,a]d[i,j]+n[i,j,a](20)
using like designations of h, d, i and j as Equation (1).

(89) The modified observation for calculating an effective LLR across all antennas can be given by a weighted sum of individual modified observation y for each antenna a in the form of Equation (9). In particular, the weights are given by the antenna noise power represented by .sub.a.sup.2:
[i,j]=.sub.ay[i,j,a]/.sub.a.sup.2=.sub.ah*[i,j,a]r[i,j,a]/.sub.a.sup.2(21)

(90) In other words, in a spatially diversified receiver, the observation modifier 222 may generate an effective modified observation as a weighted sum, based on the product of the observation received by individual antennas and the complex conjugate of the channel estimate of the channel perceived by the respective antenna, normalised by the respective antenna noise power.

(91) The noise power estimator may be configured to estimate or measure the effective SNR calculated as the sum of SNRs of individual antennas:
[i,j]=.sub.ah*[i,j,a]h[i,j,a]/.sub.a.sup.2=.sub.a|h[i,j,a].sup.2/.sub.a.sup.2(22)

(92) For a 16-QAM system, the LLR module 225 may be configured to calculate the complementary LLRs for the most significant bits (i.e. .sub.0 and .sub.2) as:
.sub.0[i,j]2.sub.I[i,j](23a)
.sub.2[i,j]2.sub.Q[i,j](23b)
where .sub.I and .sub.Q are the real and imaginary parts of the modified observation . It is noted that Equation (23a) (and hence similarly (23b)) takes the same form as Equation (16a) except that the effective modified observation is based the noise-weighted sum given by Equation (21), rather than Equation (9). Further, the LLR module 225 may be configured to calculate the complementary LLRs for the least significant bits (i.e. .sub.1 and .sub.3) as:
.sub.1[i,j]4[i,j]|.sub.0[i,j]|(23c)
.sub.3[i,j]4[i,j]|.sub.2[i,j]|(23d)

(93) It is noted that Equation (23a) (and hence similarly (23b)) takes the same form as Equation (19a) except that the effective modified observation is based on the weighted sum given by Equation (21), rather than Equation (9), and also that the SNR is the sum of channel power to noise power ratio, summed over all antennas.

(94) For a 64-QAM system, it should be apparent to a skilled person that the LLR module 225 may be configured to calculate the complementary LLRs for the most significant bits (MSB), the next significant bits (NSB) and the least significant bits (LSB) following a similar and slightly modified procedure above for 16-QAM:
For MSB: .sub.0[i,j]2.sub.I/[i,j](24a)
.sub.3[i,j]2.sub.Q[i,j](24b)
For NSB: .sub.1[i,j]4[i,j]|.sub.0[i,j]|(24c)
.sub.4[i,j]4[i,j]|.sub.3[i,j]|(24d)
For LSB: .sub.2[i,j]2[i,j]|.sub.1[i,j]|(24e)
.sub.5[i,j]2[i,j]|.sub.4|[i,j]|(24f)

(95) Again, Equations (24a) to (24d) have the same form as Equations (23a) to (23d). Equations (24e) and (24f) follow similar approximations. FIG. 9 illustrates a logic block diagram for an example of a demodulator 900 including an observation modifier 902 and a LLR module 904 for a multiple-antenna implementation in a 64-QAM system. The observation modifier 902 is configured to generate a modified observation according to Equation (21). The LLR module 904 is configured to generate LLRs according to Equations (22)-(24).

(96) The observation modifier 902 includes a first subsystem 902a for a first antenna A and a second subsystem 902b for a second antenna B. Each of the subsystems 902a and 902b is similar to the subsystem 600 FIG. 6B (without showing the saturate blocks, the inverse LUT block, the shift block or Calc exponent block for simplification), except that their outputs .sub.A and .sub.B are noise-weighted modified observations. In other words, .sub.A and .sub.B are individual terms for the respective antennas in the summation given by Equation (21). The observation modifier 902 computes the summation and provides an effective modified observation as an output. The first subsystem 902a and the second subsystem 902b, like the subsystem 600, also provides estimates of the signal-to-noise ratio for each antenna (represented by |h.sub.A|.sup.2/.sub.A.sup.2 and |h.sub.B|.sup.2/.sup.2).

(97) As shown in FIG. 9, the LLR module 904 obtains and add |h.sub.A|.sup.2/.sub.A.sup.2 and |h.sub.B|.sup.2/.sub.B.sup.2 to generate according to Equation (22). The LLR module 904 also obtains the real and imaginary parts of the effective modified observation . According to Equations 24(a)-(e), the LLR module 904 calculates LLR0, LLR2 and LLR4 (i.e. LLRs for the MSB, NSB and LSB) based on the real part and calculates LLR1, LLR3 and LLR5 (i.e. LLRs for the MSB, NSB and LSB) based on the imaginary part: In the first and the fourth branches 904a and 904d, the LLR0 and LLR1 are calculated based on Equations (24a) and (24b). For simplification, a multiplication block by a factor of 2 (according to Equations 24(a) and (b)) is not shown for LLR0 and LLR. LLR0 and LLR1 therefore is shown to take the real and imaginary parts, respectively, of the effective modified observation directly. In the second and the fifth branches 904b and 904e, the LLR2 and LLR3 are calculated based on Equations (24c) and (24d). For simplification, a multiplication block by a factor of 2 (according to Equations 24(a) and (b)) to calculate LLR0 and LLR1 is not shown. LLR2 and LLR3 are obtained by providing LLR0 and LLR1 to an absolute value block (906a and 906b) and adding to a scaled generated by the LLR module 904. The scale is 4 according to Equations (24c) and (24d). In the third and the sixth branches 904c and 904f, the LLR4 and LLR5 are calculated based on Equations (24e) and (24f). LLR4 and LLR5 are obtained by providing LLR2 and LLR3 to an absolute value block (906c and 906d) and adding to a scaled generated by the LLR module 904. The scale is 2 according to Equations (24e) and (24f).

(98) One or more of the components of the receiver 220 or the demodulator 900 may be implemented as software, such as a computer program including instructions stored in a non-transitory computer-readable medium and executable by the one or more processors. In one example, the non-transitory computer-readable medium is a memory or storage module, such as volatile memory including a random access memory (RAM), non-volatile memory including read-only memory (ROM), or a harddisk. The one or more processors may be one or more computer processing units (CPUs). Alternatively or additionally the one or more of the components of the receiver 220 may be implemented as hardware, such as using one or more digital signal processors (DSPs), application-specific integrated circuits (ASICs) or field-programmable gate arrays (FPGAs).