CHAPTERS BLIND ITERATIVE MUD WITH ERROR CODING

This chapter gives an overview of iterative multiuser detection.Selectediterative detectors that have been proposed in literature are discussed , and their performance compared . Iterative adaptations of the blind detector are summarised. Previous work on blind iterative detectors is described. The proposed blind iterative MUD, meant for data communication applications in both ruralandurbanenvironments,is presented. Thedifferencesbetween the propose d detector and existing detectors are described, and modifications that may be made to the proposed detector are given. An analytical model of blind iterative detection, using SOYA decoders is given.

5.1 Iterative MUD 5.1.1 General

Figure 5.1 shows a simplified block diagram of an iterative CDMA decoder. The soft output CDMA decoder block is described in more detail in Section 5.1.2. Depending on the method used for the CDMA detector,feed-forward and feedback filters may need to be employed.The functioning of the interleavers and de-interleavers is described in Section 4.2.The figure only shows one interleaver,a SISOdecoder anda de-interleaver.In reality,after the received signal r haspassedthrough the softoutputCDMA decoder, itis passedthrougha bank of k parallel de- interleavers,SISO decoders, interleavers and filters(not shown)where necessary,where kis the number ofusers. Essentially,eachuser's signal has its own path of de-interleavers (denote dby 1(' ),decoders andinterleavers(denoted by71:)after thesoft-outputMUD block has processed it.

This pathofa single user's signal is what isrepresentedin Figure5.1.

r ~

Soft

Output

~

^~

^COMA j ^EJ

5150

~

d Decoder

Figure 5.1:Genera l blockdiagram of anite rative decod er

59

5.1.2 Soft-Output CDMA Decoding

Many types of SISO COMA decoders for iterative decoding have appeared in literature, the most common are based on MMSE [36-42] or IC [41-46] detectors.Iterative detectors based on the decorrelating multiuser detector [40] and single user detectors [47] have also been considere d . Due to the numerous variations in the implementatio n ofthe soft-output COMA decoder, it is not practical to give an overview of them all. A general description of the workings of the iterative detector, as shown in Figure 5.1, will be given. A brief description of selected iterative detectors, whose methods were considered to be used as the basis of this project,willbe given.

The soft output COMA decoder accepts the received signal r, and generates soft outputs, which are passed through the de-interleaver to the SISO decoder,which can be either a SOY A or MAP decoder,operating as described in Section 4.4.2. It accepts soft-outp ut decisions from the COMA decod er and outp uts its own soft-decisions, which are interleaved to be processed for the next iteration.After the required number of iterations have been completed, the hard decision d is made.Iterative decodersconverge to their final valueranging between 3 iterations [48] and 10 iterations [41].

5.1.3 Selected Iterative Detector Structures

The receiver structure proposed in [47] is similar to the one shown in Figure 5.1, where the SISO block is a MAPdecoder,and the Soft Outp ut COMAblock is a 'COMA MAP' decoder.

The differencebetween thetwo isthat a MAP decoder outpu ts message symbo lsafte raccepting coded symbols, whereas the COMA MAP decoder accepts the sprea d symbols as an init ial input - it needs to be able to take into account the COMA channel. The soft output of the COMA MAP decoderis

(5.1)

wherey is the received signal,dis the coded message, L is the number of symbo ls,K is the number ofusers, and

p(~

I

eli)

=

p(~,di)/P(eli), (5.2)

whereP(di) is the a priori information aboutthe symbol d,and is set to 0.5for thefirst iteration, for bothd,

=

+Iandeli =-1,where

(5.3)

and

. . . LK '

p(~, ~ ;- K

^{+2) =}

peE : _{' ~ ;- K}

+2)

P(~ J+I I

df-K+2). ^(5.4)

The interleavers and the MAP decoder function as describ ed In Sections 4.2 and 4.4.2 respect ively.

Figure 5.2 shows the block diagram of the proposed receiver in [45]. Here the 'APP' blocks repre sent MAP decoders,and they, along with the interleavers and de-interleavers, function as describedabove.The 'tanh' functionsgeneratesoft-decisionswith which the cancelled received signal can be calculated.Each individual decoder processes the received signal and generates soft-deci sion values ofthe transmitted coded symbols d^k[J]. Once the soft estimates

d

^k[j] of the symbols have been generated by the tanh functions,a cancelled receivedsignal for user k at timej can becalculated [45]:

rk,J = Sd - Sdk';

+

^{Z, (5.5)}

where S IS a matri x of the spreading sequences, Z IS the noise term, and

d

_U ^=[dl[l]" ..,dk_I[J],O,d k+J[j] ,···,d k[K]] ,although typically only estimates around the symbol ofinterestarerequired.

..

u APP

~0 (,)

q ^{::gg <Q}

^{Q .;;}11) ^APP

U

·0

Ii)

Cl

~⁰

I n~ ^I

^lil"

f'J".J

Figure5.2 :Blockdia gr a m of joint iter a tive decoder 1451

61

Residual interference is suppressed through the use of linear filters (part of the CDM A soft decision decoder),in the general form of

Yk[j]

=

_{W~.lk,j '} ^(5.6)

whereWk,jis the filter for the kthuser at timej. The interference canceller ignores the structure of mutual interference, and uses the spreading sequence so Wk,j = Sk ,j ' The MMSE approach uses a more sophisticated filter

(5.7)

where K k,j =S~,jDkS k,j +0'²1, D_k is a diagonal matrix of the residual power of interfering user s, and Sk,j = [s^{1.1"" ,}Sk-I,j'Sk+l.j' ''·'SK.L]' As the MMSE filter is heavily dependant on the signal-to-noise ratio, the normalisation factor, Pk

/ (1 ⁺

^Pks

_{~ .j K ~ >}

^{k.J ,} can be dropped, reducing the filter to

(5,8)

By applying the stationary inversion, ^{X( I/+I)} = x' "' -

(Mx(//) - b ),

^to

K ~:j'

a multistage filter implementation is arrived at

(1/+1) <I> (K\.

Wk,; = ¹¹ Pk,j '

where each stage in the filter consists ofa simple matrix multiplication

Kk , jwi':}

[45].

(5.9)

Figure 5.3 shows a block diagram of the receiverstructure proposed in [48]. The receiver takes thematched filter output,

~"

and generatesconditional channel probabilities

p~, 1 4, ),

which are multivariate Gaussian conditional probabilities [20]. The marginal probabilities for the kth decoder,

p~, 1 4:

^{k)) ,} are calculated by the metric generator. The single-user SISO MAP decoders accept the output from the metric generator,and outputa posteriori information back to the metric generator as a priori information for the next iteration, as described above.The decision rule for the metric generator is [48]:

p(~"Y )

~I = argmax ^{- I}

ti, p(y)

- I

= argmaxp(yti, ^{_ I}

I

~I)

(5.10)

Figure 5.3:Blockdiagr amproposedbyReedet al 1481

The block diagram in Figure 5.4 shows the system proposed in [49]. This method employs a seria llyconcatenatedconvo lutionalcode,whichconsistsof aRSC encoder,an interieaver,anda differential encoder,for each user. The receiver utilises a bank ofK matched filters to decode the COMA signal. The received signal is iterated between the RSC decoder, the differential decoder and themultiuserdetector [49].

b, ~ YI ...

Multiuser Channel b_K

~ YK

...

f--- - ••

~

I I I I

r----tol.~