WO2002098090A1

WO2002098090A1 - Method and apparatus for soft information generation

Info

Publication number: WO2002098090A1
Application number: PCT/GB2002/001182
Authority: WO
Inventors: Candido Levita; Jens Tingleff
Original assignee: Cambridge Silicon Radio Limited
Priority date: 2001-05-24
Filing date: 2002-03-18
Publication date: 2002-12-05
Also published as: GB0112657D0

Abstract

A method and apparatus is disclosed for decoding a data symbol associated with a sequence of bits. The method includes the steps of estimating each bit in the sequence of bits, measuring the reliability with which each bit in the estimated sequence of bits is estimated, responsive to at least one bit being estimated with a level of confidence lower than a predetermined level, generating an unreliability tag associated with that one bit, and outputting the estimated sequence of bits together with an unreliability tag for only each of the at least one bits which are estimated with a low level of confidence.

Description

DECODING SIGNALS

The present invention relates to a method and apparatus for decoding bit estimates from a multidimensional signal set. In part, but not exclusively, the invention relates to the process of decoding hard bit estimates together with a soft bit estimate for only the bits most likely to have been decoded unreliably.

As is well known in the art, data is produced as binary digits or bits which are pulses whose voltage amplitude determines whether they represent a 1 or a 0. These pulses are grouped together and arranged to form a code. In order to transmit this data over a radio link (or other analogue channel) the signal comprising the code must first be encoded into a form which is acceptable to the network. It is modulated to map the unique bit sequences onto analogue signals, possibly taking into account the history of the data input, with the analogue signals taking values from the symbol space.

At the receiver the signal must be decoded to restore the digital format. The task of decoding signals in receivers for digital communication systems is the process of translating (multi-dimensional or multi-level/phase) signals into symbol estimates from the modulation value space .

Various modulation processes have been developed to enable data to be transmitted in this way. The method of modulation used depends upon the desired rate of transfer of the data, the available bandwidth and the maximum acceptable bit error rate (BER) .

A problem with this type of system is that noise and other non-linearities in the system can lead to distortion in received signal levels which can lead to incorrect data being decoded at the demodulation end.

Three of the principle forms of modulation used are Amplitude Modulation (AM) , Phase Modulation (PM) and Quadrature Amplitude Modulation (QAM) . In an amplitude modulation system a 0 or 1 is transmitted using the same carrier signal frequency but having two distinct amplitudes .

Phase modulation essentially involves changes of phase of a carrier signal depending upon whether a bit 0 or bit 1 is to be transmitted. In two level phase shift keying (2PSK) the phase of the carrier is modulated by 180° depending upon whether a 0 or 1 is transmitted. In Quadrature phase shift keying (4PSK) the single frequency carrier frequency is split into 2 separate carriers of the same frequency but having a 90° phase split between them. Each of these 2 carriers can be modulated with a 180° phase change to thus provide four modulated signals. This type of modulation is also susceptible to the above-referenced problem.

Quadrature amplitude modulation (QAM) combines amplitude modulation with phase modulation. The carrier signal frequency is kept constant but is divided into phases . The number of phases may be selected according to the rate of transfer of data required. Each phase is amplitude modulated.

Figure 1 illustrates how a QAM system may be combined with the use of eight phases and four amplitudes. The sixteen points 1-16 fall on the four circles, shown as dotted lines 17, 18, 19, 20 which correspond to the four amplitude modulation levels. The eight phases of the modulation signals may be represented via lines drawn from the origin to points 1 to 16. These are the constellation points in a QAM constellation. The number of amplitude levels on each phase is designated via a number placed before the letters QAM. Figure 1 thus illustrates a 4QAM system. 4QAM corresponds to quadrature phase modulation with four levels of amplitude modulation. The 4QAM map has four circles (4PAM) each with four points (4PSK) positioned so that alternating amplitude levels use 45° offset phase.

Each of the points 1 to 16 corresponds to a transmitted symbol . That is to say a unique sequence of four binary digits (bits) which is allocated to that particular combination of phase and amplitude. For example taking point 9 this has a signal amplitude which is greater than that at points 1-4 but less than that at points 13-16. It also has a phase which is the same as that at point 3. A receiver receiving a signal having this combination of phase and amplitude would thus be able to identify the fact that the transmitter has sent this symbol . Since this symbol has a unique sequence of bits associated with it the receiver symbol can be used to reconstruct an original bit sequence. In this way the decoder translates multi- dimensional signals, (the phase and amplitude dimensions) into a symbol estimate. The signal has a real and imaginary part which are dependent upon these characteristics. This symbol estimate corresponds uniquely to a sequence of bit estimates according to the modulation scheme used.

These estimated bit sequences which are estimated in a decoder are termed hard decisions . Figure 2 illustrates the sequence of bits mapped to each point 1-16 of figure 1. The bit sequences are Gray coded so that only one digit changes between each adjacent point. In this way if an input stream of data bits includes a portion 0111 this could be encoded and transmitted via a signal having a phase and amplitude as shown via point 12 of figure 1. A receiver measuring the received signal phase and amplitude would identify point 12 and know that a 0111 had been encoded and transmitted. As an alternative to the constellation points of figures 1 and 2 a constellation including points in a square configuration can be used. Instead of falling on points of similar amplitude and phase of each point are selected so that the points fall at predetermined positions.

A problem which occurs when signals are received is that, as a result of noise or other non-linearities the phase and/or amplitude received at a receiver/decoder may be at some value other than the ideal value for the transmitted symbol . These variations may lead to a decoder incorrectly identifying a transmitted symbol and thus its associated unique sequence of bit estimated. This is particularly a problem when the integrity of the transmitted data is of paramount importance.

In the past as a part solution to this problem the output from a QAM decoder has been input into an error corrector such as a Viterbi block which has some predictive ability to estimate what the original bit sequence was. These can exploit a Maximum Likelihood Sequence Estimate (MLSE) and use reliability information from the decoder which estimates the transmitted symbol. The reliability information reflects the level of confidence with which the estimate of the bit in the estimated sequence of bits is made. The combined result of estimated bit and reliability information for that estimate is called a soft decision bit estimate. Use of soft decisions increases the tolerance to noise. However prior art solutions have conveyed reliability information about all the bits of each symbol. This has demanded a high level of data processing to be carried out in the Error Correction Unit .

It is an aim of the present invention to at least partly mitigate the above-mentioned problems.

According to a first aspect of the present invention there is provided a method of decoding a data symbol associated with a sequence of bits comprising the steps of: estimating each bit in said sequence of bits; measuring the reliability with which each bit in said estimated sequence of bits is estimated; responsive to at least one bit being estimated with a level of confidence less than a predetermined level, generating an unreliable tag associated with said one bit; and outputting said estimated sequence of bits together with an unreliable tag only for each said one bit .

Preferably said step of outputting said estimated sequence of bits further comprises: outputting a reliable tag with each bit in said estimated sequence of bits which is not output with an unreliable tag.

According to a second aspect of the present invention there is provided decoding apparatus comprising: decoding circuitry responsive to a received data symbol for providing an estimated sequence of bits associated with that symbol; reliability measuring circuitry arranged to calculate how reliable the estimate of bits in said sequence is and responsive to at least one bit being estimated with a level of confidence less than a predetermined level to generate an unreliable tag associated with said one bit; whereby for each received data symbol the decoder is arranged to output the estimated sequence of bits together with an unreliable tag only for each said one bit.

Embodiments of the present invention will now be described hereinafter by way of example only and with reference to the accompanying drawings in which:

Figure 1 illustrates how symbols can be mapped. Figure 2 illustrates an example of Gray coded bit estimates for each of the symbols of figure 1.

Figure 3 illustrates a digital communication system.

Figure 4 illustrates a 16QAM square constellation.

Figure 5 illustrates encoding and soft decision decoding.

Figure 6 illustrates a 16 QAM constellation.

Figure 7 illustrates how a symbol may lie close to an expected constellation point.

Figure 8 illustrates how a soft decision bit estimate can be established.

Figure 9 illustrates a decoder.

Figure 10 illustrates how embodiments of the present invention can improve decoder performance.

In the description like reference numerals refer to these parts.

Figure 3 illustrates some of the functional parts of a digital communication network 30. In particular to the encoding and decoding functions for a downstream physical layer. It will be understood that the present invention is not limited to such an application. The digital communication channel includes a sender end and a receiver end and may for example correspond to a base station and subscriber station respectively. At the sender end data in the form of a stream of binary bits each having a value of 0 or 1 is formed to represent some information which is to be conveyed. This stream of data is input into an encoder 31 via input 32. The encoder slices the data stream into a series of bit sequences each comprising, for example, a sequence of 4 bits. It will be understood that other numbers of bits could be used. The encoder maps each of these sequences into a symbol according to a Gray coded mapping policy. The encoder output 33 is input into a transmitter 34 from which the encoded information can be transmitted via any suitable modulation for example over air interface via an RF channel. Prior to transmission the digital signals are converted into an analogue signal via digital to analogue (D/A) converters . These analogue signals can be viewed as having a Real component signal

(Q) and an imaginary component signal (I) as is known in the art. The base station may also include other function blocks (not shown) such as sync byte inversion, convolutional encoding, interleaving and base band pulse shaping prior to transmission as is known in the art. It will be understood that if such features are used analogous units should be provided in the receiver end station in order to enable the received signals to be successfully decoded. The transmitted signal 35 from the RF channel is received by a receiver 36 which includes an analogue to digital (A/D) converter for each of the I and Q components . The outputs of these A/D converters is in the form of a sequence of bits. These sequences of bits are used to identify the position of the transmitted symbol in the signal space 42 of a square QAM constellation. This is illustrated in figure 4. The QAM constellation in figure 4 corresponds to a squared gray coded constellation. Rather than falling on the similar amplitude circles of the QAM constellation in figures 1 and 2 each of the adjacent points in the QAM constellation of figure 4 are equidistant This produces an equal density distribution of constellation points. Each point corresponds to a symbol associated with a unique gray coded bit sequence as shown. The two-dimensional signal 33 is shown in an X/Y coordinate system as I (in-phase) and Q (quadrature) components. The processing in the receiver produces a digital two- dimensional signal 37 which has I and Q components using more digits than was used for encoding in the symbol, i.e. the value space of the signal 37 contains the value space of signal 33 and a number of values around the original constellation points. The receiver 36 outputs the converted receiver signals as digital coordinates for the symbols via signal 37 to decoder 38. The decoder decodes the complex signal 37 by estimating in sequence the most likely symbols being transmitted. This is a so called hard decision. From this the decoder can establish a hard decision bit estimate for each bit in the original data sequence which was encoded in the encoder 31. The decoder also determines how reliable the bit estimates are. That is to say an assessment of the level of confidence with which the estimate is made is carried out. The decoder creates reliability information for those bits in sequence which are estimates with less than a predetermined level of confidence for example less than 100% (or total) confidence. These processes will be described in more detail hereinafter. Bit estimates which include soft information (reliability information) generated in the decoder are termed soft decision bit estimates. The decoded bits from the decoder 38 including hard level bit estimates and soft bit estimates are output via signal 39 to an error correction block 40. This is arranged to exploit the reliability information from the symbol estimates 38. The output 41 from the error correction circuitry 40 comprises a stream of data bits formed by a series of successive sequences of bits and should approximate to or tally exactly with the input stream of data input on signal 32.

Due to noise or other irregularities which might occur during transmission the signals received at the receiver from the RF channel may differ somewhat from those transmitted. This may manifest itself as a variation in signal amplitude and/or a change in phase.

As a result detected signal values may not exactly lie on a constellation point corresponding to a transmitted symbol. The decoder 38 holds details of the possible symbol characteristics in a store for example either in a lookup table or by simulating details of the algorithm used to generate the symbol map. Any method for storing details of expected symbol characteristics can be used. For each received symbol the real and imaginary parts of the signal

37 are converted into a value which can be used as an index or coordinate to locate the symbol on the constellation map of figure 4. If a perfect match, i.e. the converted values index a constellation point exactly, is made the received symbol can be identified and the unique bit sequence associated with that symbol output as a sequence of hard bit estimates. If a received symbol does not have a characteristic phase and/or amplitude, and thus real or imaginary part, which matches an expected symbol it is indicative that some error has occurred during transmission and an assessment must be made to try to identify the most likely bit estimates for that symbol. This is carried out by determining which bits in the bit estimates are most likely to be accurate estimates and those estimates which are likely to be least accurate. For at least the bit estimate which is the least likely to be correct, and thus the most unreliable, that bit is tagged as unreliable. The tag also indicates the level of unreliability i.e. the level of confidence with which the bit estimate is made. The hard decision bit estimates, that is the most likely correct bit estimates, are output to the error corrector together with soft bit estimates, that is the least likely bit estimates together with an unreliable tag. Soft bit estimates correspond to hard bit estimates, if the received symbol is perfectly matched to a symbol stored in the decoder in which case the bit sequence associated with that symbol is known. In this case the bit estimates for each bit in the estimated sequence of bits associated with the received symbol are output with a tag noting their 100% reliable status. Alternatively if the received symbol characteristics do not match a known symbol soft bit estimates can be made but for only the least reliable hard bit estimate. Alternatively soft bit estimates can be made for more bits in each sequence of bits for example for the two most unreliable bits. The most reliable bit estimate or bit estimates are merely tagged as reliable. The soft decision decoding described herein relies on the assumption that the error, i.e. the distance between the transmitted symbol and the received signal, is normally smaller that the distance between adjacent constellation points. This holds for QAM modulation schemes mentioned above at the sensitivity limits defined in the relevant industry standards. This assumption makes single-bit errors in the hard estimates the most likely error phenomenon when Gray coding is used in the transmission. Therefore, one bit (out of the several bits defining a constellation point in a multi-level modulation) is most likely to be estimated incorrectly.

Figure 5 shows an embodiment illustrating the selective addition of reliability information for a one- dimensional signal with a modulation which conveys two bits per symbol (e.g. 4PAM) . The constellation points (heavy- black circles) on the received signal axis correspond to the four combinations of input bits 'b' (given above the signal axis) as encoded. The output bit estimates ^λe' (below the signal axis) are shown as decoded. The 'e' values are either the discrete values 4-1 or -1 (for hard decisions) or a value between +1 and -1 (for soft decisions) .

For a received signal a real and imaginary part which identify a point having characteristics falling on any one of the points 45-48 of the constellation the bit sequences (in this example a two bit sequence) are known with certainty. At point 45 the first bit estimate is shown to be a 0 by reference to the probability graph 49 whilst the second bit is also known to be a 0 by reference to the probability graph 50. These probability charts 49, 50 indicate the reliability of the estimates for their respective bits. The point 51 shown by a light circle on the received signal line represents a transmitted symbol which has been in some way affected during transmission say by its phase or amplitude being affected by noise. It does not therefore fall on any constellation point 45-48. Details of these probability charts may be stored so as to be accessible to the decoder circuitry.

An assessment of the bit sequence associated with this received symbol may still be made. Referring to the probability chart 49, 50 details of which are stored in decoder 38 it will be seen that the first bit is known with certainty to be a 0. The second bit, referring to 50, is most likely to be a 0 and the curve or line in the hatched area 52 can be used to obtain soft information about the level of reliability/confidence with which the bit estimate is made. The output decoder signal 39 from the decoder 38 decoding a received symbol having the characteristics of point 51 would therefore include a hard bit estimate for bit bl of 0 and a soft bit estimate of bit b2 of 0 with a tag which includes soft information identifying b2 as the least reliable bit and the probability that that bit was correctly estimated (which probability is calculated from the probability charts 49, 50 stored in the decoder).

Whilst the probability graphs 49, 50 in figure 5 are shown as having linear dependency in the shaded regions, it will be understood that some other function such as a quadratic function could be used. An embodiment illustrating the extension into two dimensions is shown in Figure 6, with soft information being added to only one bit out of the four bits conveyed

(in 16QAM) . The bit in question is indicated by the hatching of the area where this bit is the least reliable.

In this constellation each point on the columns of constellation points shares the same two least significant bits whilst points on each row share the same two most significant bits. The 4 bit sequences for some of the points are shown. Others are omitted for the sake of clarity. A received symbol having characteristics which fall on one of the constellation points shown as heavy black circles can be decoded into its associated 4 bit sequence. For example a received symbol having the phase and amplitude characteristics, and thus the real and imaginary parts shown in Figure 6 of point 55 can be decoded into bit sequence 1010. Each of these bit estimates will be output as a hard decision bit estimate. For a received input symbol signal having the characteristics of point 56 there is no constellation point sharing the same characteristics . An estimate of the most likely bits should therefore be made in the decoder 38.

At a first step the constellation point most nearly matching the received symbol is identified. For point 56 the nearest constellation bit is that associated with the bit sequence 0100 which occurs in the third quadrant 59 of the four quadrants 57-60.

The location of point 56 is illustrated in more detail in Figure 7. The location of point 56 is determined by measuring the distance from that point to the nearest constellation point. This provides two measurements Δx and Δy. A comparison is made of these two measurements. In this example Δx < Δy. Therefore the bit estimate associated with the Δx measurement is more likely to be accurate compared with that of the Δy measurement. By virtue of the constellation points being associated with Gray coded bit sequences an estimate of the most likely bit sequence associated with the received symbol corresponding to point 56 can be made. With reference to Figure 7 a comparison of the bit sequences associated with the constellation points surrounding the point 56 can be made.

The most likely sequence which has been encoded, transmitted and is now received is 0100. The fact that the Δx measurement is smaller than Δy means that an assumption can be made that the most likely point other than that of 0100 which corresponds to the transmitted symbol is that associated with 0000. This is because the point 56 is nearer the point for 0000 than the point for 0101. A comparison can then be made of the associated bit sequences for each of the most likely symbol and next most likely symbol i.e. 0100 is compared with 0000. Three of these bits are identical in the first, third and fourth bits. A hard bit estimate can thus be output for these bits since they are substantially certain to be a 0. The second bit is detected as being the least reliable bit estimate. Soft information is generated for this bit in accordance with a level of confidence graph or other such indicator shown in more detail in figure 8. The relative distance Δy from the constellation point associated with bit sequence 0100 gives an indication of the level of confidence with which the bit can be estimated. If Δy was small the second bit is more likely to correspond to a 1 since the symbol from which it is decoded is most likely to be associated with bit sequence 0100. As Δy increases the reliability of the second bit being a 1 decreases and the likelihood of it being 0 increases. As an alternative, in certain embodiments, in addition to outputting the second bit which is the least reliable as a soft decision bit estimate, the bit which is dependent upon the ΔX measurement may additionally be output as a soft decision estimate. Referring again to figure 7 it will be seen that whilst it is highly likely i.e. ΔX is small, that point 56 corresponds to 0100, if the point does not correspond to 0100 the next most likely point after 0000 is 0101. This point shares the same first, second and third bit but the fourth bit is a 1 not a 0. This fourth bit is thus output as a soft decision bit estimate as a 0 plus reliability information. In this embodiment the first and third bit estimates will thus be hard decision bit estimates whilst the second and fourth bit estimates will have reliability information tagged with them. The soft information may be output as a separate tag comprising a plurality of bits which may be set to indicate the reliability. Alternatively the tag may comprise a variation in the voltage of the output bit in the output signal . For example in the output 39 from the decoder 38 a 0 may be output at -4 volts whilst a 1 would have a signal voltage of 4-3 volts. By varying the voltage an error correcting block subsequent to the decoder can use the received voltage to estimate the level of confidence with which the bit estimate is made. For example a -2 voltage would indicate a 0 but with less confidence than an output signal voltage of -4 volts.

The reliability of the accuracy of the bit estimate is noted in the decoder which generates a tag indicating the level of confidence with which the estimate is made. This tag is attached to the output second bit estimate as a soft decision bit estimate. The other three output bits are output with a tag indicating an identical high level of certainty as hard decision bit estimates . Error correcting circuitry can use this reliability information to further help correct the errors . The process of decoding for some embodiments thus includes the steps of being given a complex input signal (S) and calculating a hard decision estimate for all the bits (i.e. el, e2,..., ej,... ek) . The estimate ej which is least reliable is identified and reliability information for that identified bit estimate ej is produced. The estimate ej is thereafter modified to produce one soft decision estimate.

In other embodiments a soft decision estimate is produced for more than one bit estimate in each sequence of bit estimates associated with each transmitted symbol.

Figure 9 shows a decoder in accordance with an embodiment of the present invention. The decoder is suitable for use with QAM mapping and performs hard decision decoding on each dimension (i.e. both the real and imaginary components of the received signal) independently and then selects a soft metric as a function of the hard decisions and the distance from the input value to the nearest constellation point. The evenly spaced QAM constellation as shown in figure 4 has 2^K points in each of the two directions (the real and imaginary directions) mapping 2 ^2K bits. The first input which corresponds to the real component of the received signal is Ri_nput • The second input which corresponds to the imaginary component of the received signal is Iinpu * These two inputs form the two-dimensional output 37 from the receiver 36. Both inputs have the form of equation 1.1.

input = i₀ A ... ik-i fo A •■• f_e-ι 1.1

With n = k+e and i₀ the most significant bit. The signal is scaled and offset so that constellation points fall in the centre of the intervals enumerated by the value given by equation 1.2.

i = i_k-1 + 2i_k_₂ + ,k-l io 1.2

For example at 8, 24, 40 and 56 for n = 6 in 16 QAM (which has k = 2) .

These are input into blocks 80 and 81 respectively which include circuitry to extract information from these signals. These blocks 80, 81 are arranged to output the i component of the inputs Rinput/ linput as k bits i equals i_k-ι + 2i_k-2 + ... 2^k_1 i₀ these enumerate the intervals corresponding to the constellation points. Thus effectively the most significant bits operate to index which of the constellation points is the most likely t correspond to the transmitted signal. The i components are output as Ri, Ii each of which includes k-bits and is input into a respective gray coder 82, 83. These Gray coders are responsive to the input signals which index the constellation points by providing the coordinates on the axis of figure 4. The bits corresponding to those constellation points indexed by Ri and A are output via connectors 84 and 85 as hard decision bit estimates R_hard and Ihard respectively. Each of these has k bits Rhard = hi = h_n hi2 ... h_lk, I_hard = h₂ = h_2i, h₂₂, ... h_2k. The hard decision bit estimates are input into a metric calculator 86 as inputs hi and h2 and as inputs of a soft decision insertion and output block 87. The extraction blocks 80 and 81 also strip away the f component of the n bit sequences Ri_nput and li_nput to provide an input value for each point in the interval associated with each constellation point. This remainder R_rem I_rem for each input is calculated in accordance with equation 1.3.

rem = f_e-ι + 2f_e-2 + ... + 2^e_1 f₀ 1.3

It will be understood that the i bits are thus used to locate the nearest constellation point whilst the f bits are used to locate the received symbols position around this point . These values are input as absolute values into blocks 88 and 89. These absolute value blocks are arranged to calculate the respective distances from the most likely constellation point in the X and Y directions respectively. These values thus correspond to Δx and Δy as hereinabove described. The distances can be calculated according to equation 1.4.

The distances are labelled Rdi_st and Idis respectively. These two outputs are themselves each input into the metric calculator block 86 as inputs dl and d2 as well as into a comparator 90. The bit f₀ which is the most significant non-indexing bit from Ri_nput and Ii_nput is also stripped off by the extractor blocks 80 and 81 and input into the metric calculator block 86 as inputs si and s2 respectively. These act as direction indicators to indicate whether the received symbol point lies to the left or right or above or below the nearest constellation point. Comparator 90 is provided to compare the two distances which are input as a first input R_di_st having the form of an e-1 bit unsigned number and a second distance input I i_st having the term of an e-1 bit unsigned number and to output a value indicating which of the two input distances is the larger i.e. to output a 1 if R_di_st < Idi_s 0 otherwise. This output C which is the result of comparing Ri_st with Idist is input into the metric calculator block 86 as input s. The metric calculator block 86 outputs signals into a soft decision insertion and output block 87 which provides an output P which includes both hard and soft decision bit estimates. These can be presented to a following error correction block or deinterleaver . The truth table for the output of metric calculator block 86 for 16 QAM is set out below in Table 1. s=0, h2=xx s=l, hl=xx ll l2 si I 21h22 s2 10 0 0 4

0 0 0 4

0 0 1 1 0 0 1 3 0 1 0 1 0 1 0 3

0 1 1 0 0 1 1 2 1 1 0 0 1 1 0 2 1 1 1 1 1 1 1 3 1 0 0 1 1 0 0 3 1 0 1 4 1 0 1 4

TABLE 1 .

The output i represents the output estimate and is the index of the bit which receives soft decision information. A reserved value of the output I corresponds to no insertion, i.e. to complete confidence; this value is greater than the largest possible address generated for a proper insertion, i.e. the reserved value is greater than 2k (shown as 4 in Table 1) . This is input into the soft decision insertion and output block 87 as input a.

The metric calculator block also calculates the soft decision information SOFT which is output as output v into input d of the output block 87. The soft decision information is calculated from the largest of the two distance values dl and d2 when s = O, v = SOFT (dl) and when s = 1, v = SOFT (d2) . The output v which has m bits and is the soft information is generated according to the truth table shown in table 2.

TABLE 2

In this way the soft decision insertion and output block 87 receives the two k bit numbers Rhard and Ihar from the Gray coders 82 and 83 as inputs hi and h2 respectively. These numbers are generated in the Gray coders 82 and 83 according to the truth table of table 3.

io = = 0 io = 0 Oi = = 0 = 0

A = = 0 = 1 Oi = = 0 o₂ = 1 = = 1 A = 0 Oi = = 1 0₂ = 1

A = = 1 = 1 Oi = = 1 ₂ = 0

TABLE 3

This indicates the i₀ and A index bits used to index the constellation points and the outputs Oi o for 16 QAM (i.e. k = 2) using ETSI HIPERLAN type 2 bit numbering and encoding.. The output p consists of 2 k (m + 1) bits as 2 k output words of length m + 1 which are of the form listed below in equation 1.5.

'0,0 0,1 0,2 ^■ 0,m 1,0 1,1 1,2 fl,m

P2 -1,0 P2 -l,l P2k-1,2 ■ • • P2k-l,m 1-5

An example coding would be a two's complement signed number with values between -(2^m) corresponding to fully reliable 0 through 0 for completely unreliable to 2^m~1 for fully reliable 1 as shown in equation 1.6.

Po.o = 1 - n, o,ι = p₀,2 = i-π Pi.o = 1 - -h₁₂,Pι,ι = Pι,2 = h₁₂

p₃,o = 1 - -h₂₂,P3,ι = P3,2 = -h₂₂ 1.6

For 16 QAM (with the remaining bits set to 0) . For this encoding the soft information is inserted (if the address is in range) taking into account the hard decision i.e. viewing the last m bits as an unsigned number:

Embodiments of the present invention provide the advantage over systems where soft decisions are made for all bit estimates and/or where all dimensions of the received signal contribute to the soft decision estimate that the complexity of the decoding circuit is only modestly increased compared to a pure hard decision decoding circuit . In addition the performance of the most likely sequence estimator can be improved by the absence of information which is irrelevant. In other words soft decision estimates for bits which are more reliable than the least reliable bit or a predetermined level of confidence can be omitted. Another advantage can be seen by referring to Figure 10 which compares the effectiveness of the present invention in comparison to other decoders. Embodiments of the present invention allow a Viterbi decoder to operate with an identical output error rate 2 x 10⁴ at a signal to noise ratio which is 1.5 dB lower than in an AWGN channel with 16 QAM modulation. The improved performance is even greater for 64 QAM.

As an alternative embodiments of the present invention can also be provided to remove reliability information from soft decision decoded bit estimates. This is for modulation schemes where it is possible to assess the reliability of the individual bits as a function of the received signal and knowledge about the mapping of bits onto symbols. An example of a decoding system where this inverse sequence of operations is practical is the decoding of QPSK by assessment of the real and the imaginary part of the signal individually. Soft decision estimates can thereafter be estimated individually and then compared. Soft estimates can be changed from soft to hard for the more reliable bit.

In embodiments of the present invention the number of soft decision estimates inserted into the result of decoding a multi-dimensional signal can be 1.

Alternatively this number can be greater. Larger numbers of soft decision estimates can be inserted by treating each dimension separately. For example by treating 16 QAM as two independent 4 PAM signals and producing two soft and two hard decision estimates for each input value. Embodiments may also provide for dynamic selection of the number of soft decision estimates which are inserted by observing the range of unreliability of bits and inserting several soft decision estimates if several bits have high unreliability.

Embodiments of the present invention can be used in systems which do not use modulation with memory between symbols (typically because of re-ordering of bits before transmission) and use a mapping between bits and symbol values which minimises the number of bit erroneously decoded for a given error (ie Gray coding) .

An example of modulation schemes with the properties described above is OFDM systems where the individual sub- carriers carry phase and/or amplitude modulated signals (eg QPSK or QAM) , such as the modulation schemes of the ETSI HIPERLA Type 2 and ETSI DVB-T standards. HIPERLAN Type 2 is a flexible radio LAN standard designed to provide high speed access (upto 54 M bits per second) to a variety of networks including 3G mobile core networks, ATM networks and IP based networks and also for private use as a wireless LAN system. Basic applications include data, voice and video, with specific QoS parameters taken into account .

Embodiments of the present invention solve the problem of conveying reliability information about all the bits of each symbol by assigning this reliability information to only a single bit estimate. One soft decision bit estimate and k- 1 hard decision bit estimates are generated for a modulation scheme where k bits are mapped onto each symbol. In the case of modulation schemes with one bit per symbol, eg BPSK, only soft decision estimates are generated. This decoding operation can be done in one or more dimensions. For example, to create either 2 soft and 4 hard bit estimates or 1 soft and 5 hard bit estimates for 64QAM. In multi -dimensional modulation schemes with different signalling for different dimensions (eg combined Amplitude and Phase Shift Keying modulation) use of the described approach can be made either individually in each dimension or in combination.

It will be understood by those skilled in the art that the present invention is not restricted to the above mentioned examples but rather modifications could be made without departing from the scope of the invention.

Although embodiments have been described in relation to receivers for digital communication systems it will also be understood that the invention is not so restricted.

Rather the invention could be used wherever decoding of an encoded signal is required.

Claims

CLAIMS :

1. A method of decoding a data symbol associated with a sequence of bits comprising the steps of : estimating each bit in said sequence of bits; measuring the reliability with which each bit in said estimated sequence of bits is estimated; responsive to at least one bit being estimated with a level of confidence less than a predetermined level, generating an unreliability tag associated with said one bit; and outputting said estimated sequence of bits together with an unreliability tag only for each said one bit.

2. The method as claimed in claim 1 wherein said step of outputting said estimated sequence of bits further comprises : outputting a reliability tag with each bit in said estimated sequence of bits which is not output with an unreliability tag.

3. The method as claimed in claim 2 further comprising the steps of : for each bit in said estimated sequence of bits which is estimated with a level of confidence at least equal to said predetermined level of confidence, generating a reliability tag associated with that bit; and outputting this reliability tag with said output estimated bit sequence.

4. The method as claimed in any one of claims 1 to 3 further comprising the steps of: identifying the bit in said estimated sequence of bits which is estimated with the least level of confidence and outputting said estimated sequence of bits with an unreliability tag for only that identified bit.

5. The method as claimed in any one of claims 1 to 3 further comprising the steps of: identifying the two bits in said estimated sequence of bits which are estimated with the least level of confidence and outputting said estimated sequence of bits with an unreliability tag for only those two identified bits.

6. The method as claimed in claim 1 further comprising the steps of : for each bit in said estimated sequence of bits measuring the level of confidence with which each said estimate is made and providing an unreliability tag indicating this level of confidence for each estimated bit; and subsequently for each estimated bit which is estimated with a level of confidence at least equal to said predetermined level, converting said unreliability tag to a reliability tag.

7. The method as claimed in any preceding claim further comprising the steps of: prior to said step of estimating each bit in said sequence of bits converting said data symbol from a multidimensional signal into a digital format.

8. The method as claimed in claim 7 wherein said symbol is converted, via at least one analogue to digital converter, into a first and second digital signal corresponding to the real and imaginary parts of said data symbol respectively.

9. The method as claimed in claim 8 further comprising the steps of : comparing a portion of each of said first and second signals with data stored in decoding circuitry and responsive thereto locating a most likely symbol estimate having associated with it a unique sequence of bits.

10. The method as claimed in claim 9 further comprising the steps of : providing this unique sequence of bits as said estimated sequence of bits.

11. The method as claimed in any one of claims 8 to 10 further comprising the steps of: via a further portion of each of said first and second signals identifying how characteristics of said data symbol differ from a most likely symbol estimate; and responsive to said difference calculating a level of confidence for the estimate bit in said sequence of bits.

12. Decoding apparatus comprising: decoding circuitry responsive to a received data symbol for providing an estimated sequence of bits associated with that symbol; reliability measuring circuitry arranged to calculate how reliable the estimate of bits in said sequence is and responsive to at least one bit being estimated with a level of confidence less than a predetermined level to generate an unreliability tag associated with said one bit; whereby for each received data symbol the decoder is arranged to output the estimated sequence of bits together with an unreliability tag only for each said one bit.

13. The decoding apparatus as claimed in claim 12 where said unreliable tag indicates that said one bit associated with the unreliability tag is estimated with a level of confidence below said predetermined level .

14. The decoding apparatus as claimed in claim 12 or 13 wherein said unreliability tag indicates the level of confidence with which said one bit associated with said unreliability tag is estimated.

15. The decoding apparatus as claimed in claims 12 to 14 wherein the reliability measuring circuitry is also arranged to generate reliability tags which are tagged to bits estimated with a level of confidence above said predetermined level .

16. The decoding apparatus as claimed in any one of claims 12 to 15 wherein the unreliability tag is provided for only one bit in the estimated series of bits of each received symbol .

17. The decoding apparatus as claimed in any one of claims 12 to 16 wherein said a received data symbol comprises one of a plurality of data symbols received by the decoder each of which is associated with a unique sequence of encoded data bits.

18. The decoding apparatus as claimed in any one of claims 12 to 17, wherein the decoding circuitry further comprises: storage means including details of expected symbols each corresponding to a unique sequence of bits; and comparison means arranged to compare each of the received symbols with said details of the expected symbols and to determine for each received symbol which expected symbol most closely matches that received symbol .