Transmit Diversity in 3G CDMA Systems


Published on

Published in: Technology, Business
1 Like
  • Be the first to comment

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Transmit Diversity in 3G CDMA Systems

  1. 1. WIDEBAND WIRELESS ACCESS TECHNOLOGIES TO BROADBAND INTERNET Transmit Diversity in 3G CDMA Systems R. Thomas Derryberry, Steven D. Gray, D. Mihai Ionescu, Giridhar Mandyam, and Balaji Raghothaman, Nokia Research Center ABSTRACT Multiple antennas can improve the perfor- mance of a wireless communication system in a Transmit diversity (TD) is one of the key fading environment [1]. Although multiple contributing technologies to defining the ITU antennas may be employed at either the base endorsed 3G systems W-CDMA and cdma2000. station, mobile station, or both, it is most cost Spatial diversity is introduced into the signal by effective and practical to employ multiple anten- transmitting through multiple antennas. The nas at the base station. Hence, the topic matter antennas are spaced far enough apart that the of this article is restricted to the case of employ- signals emanating from them can be assumed to ing multiple antennas at the base station. undergo independent fading. In addition to The spacing of the antennas also affects the diversity gain, antenna gain can also be incorpo- degree of correlation between the channels from rated through channel state feedback. This leads the antennas to the mobile. Large antenna spac- to the categorization of TD methods into open ing, on the order of several carrier wavelengths, loop and closed loop methods. Several methods leads to uncorrelated fading, which leads to max- of transmit diversity in the forward link have imum performance gain due to spatial diversity. been either under consideration or adopted for Beamforming methods, on the other hand, uti- the various 3G standards. This article describes lize antenna spacing less than the carrier wave- the concept of transmit diversity and explains length, typically half the wavelength. the features of selected TD techniques. The rest of this article is organized as follows. We provide the reader with an introductory overview of diversity in general. We describe the INTRODUCTION different classes of TD and make summary remarks. The World Wide Web and increasing demand for wireless services (e.g., voice and data) are driving TRANSMIT DIVERSITY BASICS the demand for increased system capacity, data rates, and multimedia services. The International THE CHANNEL Mobile Telecommunications in 2000 (IMT-2000) Most mobile communication channels must com- standards development process, within the Interna- bat the effects of fading caused by multipath prop- tional Telecommunication Union (ITU), is driving agation. An important way of quantifying fading is the development of enhanced third-generation in terms of a measure called the coherence band- (3G) standards in order to address current and width which indicates the amount of bandwidth future wireless service needs. Particularly the Third that will fade in a correlated fashion at any instant Generation Partnership Project (3GPP) and Third in time. To define this correlation, consider a lin- Generation Partnership Project Two (3GPP2) are ear model of a communication channel; Fig. 1a developing the wideband code-division multiple illustrates what is termed the delay spread of the access (WCDMA) technologies and CDMA2000, channel. Figure 1 offers a model where the multi- respectively. Improvement of downlink capacity is path arrivals decrease in power as a function of a one of the main challenges facing the effort toward discrete time index and Td is the maximum dura- 3G evolution. Many of the proposed services are tion of the mobile communication channel. The expected to be downlink-intensive, and moreover time index is a measure of the time of arrival rela- likely to be used in low-mobility environments tive to the first multipath component at time 0. under single-path conditions. Poor performance Often, the “direct path” arrives first, and subse- due to prolonged deep fading of the channel is one quent paths represent paths reflected at increasing of the problems associated with this model. Trans- distances from the receiver. Given Fig. 1, the mit diversity (TD) is one of the key contributing coherence bandwidth is approximated by (typically technologies to addressing this problem in these path powers less than 5–10 percent of the total proposed 3G CDMA systems. power are ignored) 68 0163-6804/02/$17.00 © 2002 IEEE IEEE Communications Magazine • April 2002
  2. 2. 10 10 5 5 0 0 Multipath magnitude –5 –5 Amplitude (dB) Amplitude (dB) –10 –10 –15 –15 –20 –20 –25 –25 –30 –30 –35 –35 Ts 2Ts 3Ts Td 0 1.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4 Delay(s) Time (s) Time (s) I Figure 1. a) Delay profile; b) single-path envelope; c) two-path envelope. waveform at the receiver, where the chance that 1 all copies are simultaneously in a fade is very Bc ª . Td small. Common methods of TD employing spa- tially separated antennas utilize either temporal Considering a communication system with or frequency techniques, or combinations of bandwidth Bw, if Bc > Bw, the channel between these techniques. the transmitter and receiver is called a flat fading channel and if B c < B w the channel is called a TEMPORAL (DELAY) DIVERSITY frequency selective channel. Delay diversity for two antennas, shown in Fig. Flat fading channels are problematic for sys- 2, is a simple TD scheme that helps combat flat tems without TD, because a deep fade can result fading. Bits in Fig. 2 are generated by a source in a received signal that is below the background consisting of information from a computer, a noise level, making communication unreliable. digitized speech signal, or after being encoded The worst type of channel conditions for many by a channel encoder. The bits are numbered communication systems are slowly changing flat such that a bit at time instant n is denoted b[n]. fading channels; this is due to the length of time The original bits are transmitted using two the receiver cannot reliably demodulate the bits antennas, where the first antenna transmits with- sent by the transmitter. Using a simple model of out delay and the second sends b[n] after a delay the complex baseband communication signal of one or more sample instants. The resulting S(t), the signal at a receiver from a flat fading waveform at the input to the receiver is channel is given by X d (t ) X(t) = a(t) S(t) + g(t), = a1 (t )Â b[n]w(t - nT ) + a2 (t )Â b[n - 1]w(t - nT ) + g (t ) where a(t) represents the channel coefficient n n subject to fading, and g(t) is an additive noise = Â b[n]{a1 (t )w(t - nT ) + a2 (t )w(t -(n + 1)T )} + g (t ), process. Figure 1b offers an example of a flat n Rayleigh fading channel where a(t) is a complex Gaussian random process and Ía(t)Í is Rayleigh where ak is the fading coefficient for an indepen- distributed. When ÚÍa(t)Í 2 dt < ÚÍg(t)Í 2 dt, the dent flat fading channels, w(t) is the modulating strength of the communication signal is less than waveform for each bit, and T is the amount of the background noise, making it difficult, in time each bit is transmitted before moving to the many cases, to recover S(t). next bit. The effect of delay diversity on a slowly Transmit diversity can improve the receiver fading channel is to allow the receiver to coher- performance in the presence of flat fading. It ently add the two independent fading channels reduces the impact of fading by offering multiple together to aid in demodulation. Typically, independent copies of the digitally modulated unique pilot symbols are sent on each antenna, Base station Base station b(n) a1(t) b(n) a1(t) Convol. w(t) Convol. ej2πpfc2t encode encode ej2πpfc1t a2(t) a2(t) Delay (a) Delay diversity (b) Frequency diversity I Figure 2. Delay diversity and frequency diversity. IEEE Communications Magazine • April 2002 69
  3. 3. OPEN LOOP xe TRANSMIT DIVERSITY IN 3G b[n] h1(t) In open loop diversity methods, a predetermined Convol. encode Demux W(t) p1(t) form of diversity is introduced using multiple h2(t) antennas. Advantages of this class of methods xo include: • Signaling overhead is not required to achieve this form of diversity. W(t) p2(t) • The mobile station (MS) receiver complexity is kept relatively low. I Figure 3. OTD transmitter. The most obvious disadvantage is that the chan- nel environment information is not utilized; that is, open loop techniques are a one-size fits all allowing the receiver to characterize the two approach to achieving TD for all mobile users. channels formed between each antenna and the The earliest open loop diversity techniques mobile. Considering a case where a1(t) and a2(t) were simple in their configuration, for example, are identically distributed complex Gaussian ran- phase-switched TD (PSTD) and time-switched TD dom processes, Fig. 1c shows the response of (TSTD). PSTD introduces a known periodically a(t) =(Ía 1 (t)Í 2 + Ía 2 (t)Í 2 ) 1/2 . The fade depth, varying phase difference between the symbols difference between the peaks and valleys, is less transmitted through different antennas to simu- in Fig. 1c than that experienced in Fig. 1b. Thus, late fast fading. In TSTD the transmission is the resultant channel is more reliable from a switched among the different antennas with a communication perspective. known periodicity. All antennas transmit the This approach suffers from reduced through- same symbol simultaneously at reduced power, put due to multiple transmissions of the same so the total power remains unchanged. Each of symbol over time. Another instance of temporal these methods has been proposed at one time or diversity may be achieved in multipath channels another in the 3G CDMA standards bodies. where the signal bandwidth is larger than the TSTD was adopted for use on the synchroniza- coherence bandwidth of the channel; in this case tion channel in 3GPP. However, PSTD was not the multipaths are resolvable and may be recov- adopted in favor of other techniques such as ered by a rake receiver. Frequency diversity orthogonal TD (OTD) [2], space-time TD (STTD) methods similarly can improve the receiver per- [3], and space-time spreading (STS) [2]. formance in the presence of flat fading. ORTHOGONAL TRANSMIT DIVERSITY FREQUENCY DIVERSITY Orthogonal TD [2] is an open loop method in Frequency diversity methods (Fig. 2b) employ which coded interleaved symbols are split into transmission of multiple symbol replicas over even and odd symbol streams and transmitted multiple carriers, each separated in frequency by using two different Walsh codes. The length of a sufficiently large amount to ensure independent the Walsh code is doubled so that the total num- fading. To ensure independent fading employing ber of Walsh codes available is not reduced as a this technique, the difference between the two result of splitting the data, and the data rate will carriers, f c1 and f c2 , must be greater than the remain more constant than is the case with no coherence bandwidth (i.e., Ífc1 – fc2Í ≥ Bc). data splitting. Consider the two-antenna case. Using notation as described in the previous Let x o and x e be the odd and even symbols, section, the resulting waveform at the input to respectively. Then the symbols transmitted over the receiver is the two antennas, S1 and S2, are given by X d (t) S1 = xeW, j 2 pf c1 ( t - nT ) j 2 pf c2 ( t - nT ) — = a1 ( t)Â b[ n] e + a2 ( t)Â b[ n] e S2 = xo W, n n — + g (t) where W, W are complementary Walsh codes used (same chip rate, covering twice as many n { = Â b[ n] a1 ( t) e j 2 pf c1 ( t - nT ) + a2 ( t) e j 2 pf c2 ( t - nT ) ¸ + g ( t). ˝ ˛ chips as in the absence of OTD, but in the same number). The signal received at the mobile receiver will be Similar to TD, the effect of frequency diversity for a slowly fading channel is to allow the receiver r = h1s1 + h2s2 + g, to coherently add the two independent fading where h 1 , h 2 are the channels from the two channels together to aid in demodulation. This antennas to the MS, and p1(t) and p2(t) are the approach is accompanied by the additional cost of antenna-specific pilot signals, as shown in Fig. 3. increased complexity at both the transmitter and The time subscripts have been left out for brevi- receiver, along with the fact that it may be diffi- ty. The received signal from the two antennas is cult to implement in bandwidth-limited systems. despread using the same Walsh codes, and then Given this brief overview of TD basics, our atten- combined to recover the original symbol stream. tion focuses more specifically on the issues of TD in the context of 3G CDMA evolution. TRANSMIT DIVERSITY VIA SPACE-TIME CODING Several methods of TD have been proposed for 3G CDMA evolution. These can be broadly cate- Space-time coding is a means of enhancing the gorized into open loop and closed loop techniques. level of diversity presented to a receiver in a 70 IEEE Communications Magazine • April 2002
  4. 4. wireless link, via the addition of TD and in order takes advantage of the uncorrelated fading to more efficiently combat the signal fading across the L transmit antennas without incurring STTD is an inherent to wireless communication channels. any bandwidth expansion. Motivated by the information-theoretic results In the case of full rate transmission, L = l. In open-loop by Foschini and Gans [4] and Telatar [5], early this situation, an orthogonality property for the technique in ideas on TD schemes (e.g., delay diversity, in square space-time block code matrices [6], allows which a second antenna transmits a delayed easy recovery of the symbols arriving from differ- which the replica of another transmit antenna’s signal) ent transmit antennas despite their superposition have been refined by the work of Tarokh et al. (in time) at the receiver’s input. For complex symbols are [6]. Since it is advantageous to separate the modulator constellations the only known rate one modulated using problem of combating fades from that of chan- constructions are 2 ¥ 2 (i.e., for two transmit nel equalization, the criteria for designing space- antennas). The construction for two transmit the technique time codes are usually derived in the context of antennas was first proposed by Alamouti in [3] described in. narrowband modulation and frequency nonselec- and is defined by the simple 2 ¥ 2 pattern, tive fading. The noteworthy fact about this This type of approach is that it isolates TD from those forms È xo xe ˘ of diversity associated with the radio channel Í * * ˙, open-loop (e.g., due to multipath). Nevertheless, spread Î- x e x o ˚ Í ˙ transmit diversity spectrum systems in frequency selective channels can benefit equally from coding with space and where xo, xe are valid complex symbols from the has been adopted time redundancy, as outlined below. signal constellation (same on both antennas). by the 3GPP, due In general, coding with space and time redun- Matrices like this are unitary, cover two symbol dancy is accomplished by finding an efficient way epochs, and allow easy recovery of x o, x e at the to the fact that to allocate different symbols to different anten- receiver given the channel state [3, 6]. Alam- nas while adding, jointly across antennas, some outi’s idea, based on the Hurwitz-Radon trans- this type of type of time redundancy for implementing for- form, was further refined by Tarokh et al. [6]. transformation ward error correction. For each of the symbol streams associated with different antennas, the Space-Time Transmit Diversity — STTD is an maximizes system can then resort to other means to combat open loop technique in which the symbols are diversity gain. frequency selective fading. For example, orthog- modulated using the technique described in [3]. onal frequency-division multiplexing (OFDM) This type of open loop TD has been adopted by naturally lends itself to being used in conjunc- the 3GPP because this type of transformation tion with TD; likewise, when the excess delay is maximizes diversity gain. small, space-time block coding (see below) can STTD is defined for two antennas. Assume easily be used in a maximal ratio combining once again that x o and x e are the odd and even receiver for frequency selective channels. symbols, respectively. Then the transmissions Space-time coding can be implemented in over the two antennas, s1 and s2 are given by either block [3, 6], or trellis form [7]. Irrespec- tive of form, transmission over L transmit anten- s1e = xoW , nas can be represented by a code matrix, s2 e = xeW , È c( 1 ) s1o = - x*W , K ck ˘ ( L) e c( 2 ) Í k k ˙ s2 o = x*W , o Í c( 1 ) c(2+)1 K c( L ) ˙ Dc = Í k + 1 k k + 1 ˙, Í M M O M ˙ where W is the orthogonal Walsh code used Í (1 ) (Fig. 4). (2 ) K c( L ) ˙ Î ck + l - 1 ck + l - 1 k + l -1 ˚ The received symbol is decoded over two consecutive time epochs. The received symbol where the columns represent antennas and the may be represented in vector form as rows correspond to modulator symbol epochs; here, c (1) is the complex symbol, transmitted at n È re ˘ È h1 xoW + h2 xeW ˘ È g e ˘ symbol epoch n, from the modulator constellation Í ˙=Í ˙ + Í ˙. used on the ith transmit antenna, and c refers to Í ro ˙ Í- h1 x*W - h2 x*W ˙ Í g o ˙ Î ˚ Î e o ˚ Î ˚ the vector obtained by reading D c row-wise. A code matrix covers l symbol epochs, starting with Neglecting the Walsh codes, an estimate of the kth symbol and ending with the one indexed the transmitted symbols may be formed as by (k + l – 1); here, l is a meaningful number of epochs. For example, in a trellis-based implemen- È ˆe ˘ È h * r - h r* ˘ x tation, l could cover a codeword or frame forced Í ˙ = Í 2 e 1 o ˙. to start and end in the zeroth state; in a block Í Î ˆo ˚ Í h * r + h r * ˙ x ˙ Î 1 e 2 o˚ space-time code, l spans a block of symbols that are processed together during detection [6]. The STTD scheme is particularly simple, in Space-time block codes of rate one are based the sense that it implements Alamouti’s space- on constructing code matrices of size L ¥ L, such time block code (2 ¥ 2 code matrices, see above) that each complex symbol (arising from a group and follows it by separate spreading and scram- of encoder output symbols after mapping to the bling, as in the nondiversity mode. The orthogo- relevant modulator constellation) is transmitted nality property of the code matrices allows the by any one antenna only once (possibly complex symbols from the two transmit antennas to be conjugated and/or scaled by ±1, ±j; here j denotes — separated at the receiver front-end. There is no ÷–1). In effect, this implements a modulator that need for separate Walsh codes on the two trans- IEEE Communications Magazine • April 2002 71
  5. 5. Now ^e, ^o are concatenated and input to the x x STTD decoder for demodulation. We stress that the xe transform advantage of STS over OTD is that all symbols are xo xe transmitted over all antennas; hence, it provides b[n] h1(t) the addition of temporal diversity in the form of Convol. repetition coding prior to the decoding process. Demux W(t) p1(t) encode –x*e x*o h2(t) SCHEMES FOR MORE THAN TWO ANTENNAS xo Theoretically, the number of antenna elements through which independent channels can be W(t) p2(t) transmitted bound the achievable order of spa- tial diversity. A few open loop schemes have I Figure 4. STTD transmitter. been proposed for four antennas: • A concatenation of the OTD scheme men- tioned earlier and the STS scheme has been mit antennas for the traffic channel because the proposed as a diversity technique using four orthogonality between space-time code matrices antennas [8]. is realized in the time domain, just as in fre- • An extension of the Alamouti scheme in an quency nonselective fading. However, separate earlier section for three or four antennas Walsh codes are needed for the antenna pilot called ABBA has been proposed [9]. It has signals in order to distinguish the channels. been proven that orthogonal designs do not exist for complex channels for four antennas. Space-Time Spreading — STS [2] is another Hence, this is a suboptimal construction, open-loop technique in which the symbols are which involves some interference cancella- spread using multiple Walsh codes. It differs tion along with space-time decoding. slightly from STTD, as explained below. Of course, apart from Walsh spreading, the symbols are spread by a long spreading code, but this will CLOSED LOOP be self-understood and omitted here for simplic- ity. The differences from STTD arise in the need TRANSMIT DIVERSITY IN 3G for STS to be compatible with certain details of Closed loop diversity techniques are adaptive in the IS-2000 specifications, in particular OTD. nature. The BS obtains knowledge of the down- This was not the case within the 3GPP standard, link channel from the MS via feedback signaling, which made the implementation of STTD much and uses this knowledge to its advantage. The more straightforward. use of feedback in transmit antenna arrays was Using similar notation as in an earlier sec- first proposed by Gerlach and Paulraj [10] as tion, the symbols transmitted over the two anten- transmit beamforming. They proposed that train- nas are ing signals be transmitted periodically on the downlink and the responses of the various MSs s1 = xoW - x*W , fed back to the BS. This information is used to e calculate the optimal transmit weights for each s2 = xeW + x*W , o mobile such that the received power at the desired MS is maximized and interference to where (.)* stands for the conjugate operator. STS is other MSs is minimized. These TD techniques another simple implementation of Alamouti’s con- can be described as customized to fit the channel struction [3], based on the Hurwitz-Radon trans- — conditions for each mobile user. form [6]. If one views W, W as playing the roles of As explained at the beginning, the goal of the two transmit antennas, the Alamouti pattern in inducing diversity runs somewhat contrary to terms of xe, xo is easily recognizable; this is no sur- — that of inducing directionality using beamform- prise since W, W are, de facto, associated with the ing in that the antennas have to be spaced far two antennas. The trick is that although the sym- apart. But the problem formulation for calculat- bols in the even and odd streams completely over- ing the antenna weights remains the same if one lap in time (just as in OTD), they are recognizes the fact that knowledge of the differ- distinguishable due to spreading by the orthogonal — ent channel coefficients is equivalent to knowl- Walsh codes W, W. In other words, we do not need edge of the directional array manifold vector in two symbol epochs to implement the orthogonal the case of beamforming. In this sense, the space-time block pattern; orthogonality of two dis- closed loop diversity techniques considered in joint time epochs has been replaced by orthogonali- the 3G standards are variants of the approach in ty in the spreading code domain. The result is that [10]. In fact, correlated fading models for multi- any symbol in both the even and odd streams is ple antennas and closed loop solutions for the exposed to both fading channels, thereby experienc- same have been considered recently in these ing diversity due to lack of fading correlation across fora, arising when operators are constrained by transmit antennas and independent of interleaving. considerations of space from placing antennas The recovery of the symbol stream is as close to each other at the BS. shown below: SWITCHED TRANSMIT DIVERSITY ˆo = rWh1 + ( rWh2 )* , x * * Switched TD (STD) is an extension of the open loop technique, TSTD. In this scheme, the symbols ˆe = rWh2 - ( rWh1 )* . x * * are transmitted over one antenna at any given time. The MS uses the average received power from the 72 IEEE Communications Magazine • April 2002
  6. 6. common pilots from each antenna, and makes a decision as to from which antenna it would like the Base station BS to transmit. This decision is then conveyed to the BS through a feedback channel. This technique has been proposed in the 3G CDMA standards b[n] h1(t) Convol. bodies, but a more general and aggressive form of encode W1(t) p1(t) STD was adopted by 3GPP: TXAA. h2(t) TRANSMIT ADAPTIVE ARRAY V(t) Transmit adaptive array (TXAA) is a technique in which the MS periodically sends quantized esti- W2(t) p2(t) mates of the optimal transmit weights to the BS via a feedback channel The transmitter weights Mobile station are optimized to deliver maximum power to the MS. Figure 5 depicts the concept of TXAA. Proceeding into more detail, consider a chan- Estimate Demod nel model with a single path channel emanating W1[l]and W2[l] b[n] from each of the two BS antennas denoted h1(t), from pilots h2(t) and depicted in Fig. 5. The discussion can also easily be extended to the case of M anten- Send nas (M > 2). Since artificially induced diversity W1[l] and W2[l] is most advantageous in the case of flat fading, to BS we will consider the one path case here. Results can also be demonstrated for multipath chan- I Figure 5. Transmit adaptive array method. nels. Let the transmitter antenna weights for the current instant be w 1 [l], w 2 [l]. Let b[n] be the data symbol at the current instant and v(t) the an earlier section it can be shown that the maxi- user’s specific spreading sequence. The discrete mum achievable SNR of STTD after channel time subscripts on w and b are different since estimation is their periodicities are different. We assume that the paths from the two antennas are so closely 2 2 h1 + h2 Es spaced in time of arrival at the MS that they are SNR £ . indistinguishable. Ignoring the time subscripts, 2 N0 the signal received at the MS will be Clearly, the maximum SNR of STTD cannot Èw ˘ be greater than the maximum SNR of TXAA. [ ] y = h1 h2 Í 1 ˙ b + g Í w2 ˙ Î ˚ Details of TXAA may be found in [11] and its associated references. = hwb + g , ISSUES AND SOLUTIONS where g refers to the additive noise. In order to Precision — Under the ideal conditions of infi- maximize the received signal power, the optimal nite precision instantaneous feedback, closed transmit weights are given by w = hH/hhH. loop schemes with feedback offer a substantial The weights are normalized so that the total performance advantage over schemes without transmitted power is not altered. In the case of feedback under slow flat fading conditions. How- multipath channels emanating from each antenna ever, several issues arise in the practical imple- (if h were a matrix instead of a vector), the opti- mentation of these schemes. Limited availability mal weights will be given by the principal eigen- of feedback capacity makes the precision of the vector of the channel correlation matrix hHh. feedback an important factor. In fact, in Thus, the MS calculates the weights at peri- WCDMA, a feedback capacity of 1500 b/s is odic intervals from the information h obtained assumed, which amounts to 1 b/slot. Several through the two strong pilot signals P 1 and P 2. methods have been used to convey channel These weights are quantized and then fed back information at this bit rate: to the BS on the reverse link control channel. It • Quantize the complex feedback coefficient is also worth noting that the STD method to 1 bit of magnitude and 3 bits of phase described previously is actually a subset of and send them over successive slots [12]. TXAA, with the weights being [0 1] or [1 0]. • Feedback only the phase information for If one assumes that the feedback mechanism in the complex coefficients. Set partitioning is TXAA perfectly tracks the channel conditions of done on the phase constellation, and the the downlink, the signal-to-noise ratio (SNR) after transmit weighting is calculated by filtering demodulation and channel estimation is bounded as over multiple feedback bits [12]. 2 Ê ˆ Feedback Error — The feedback bits are not 2 2 Á h1 + h2 ˜ Es Ê 2 2 ˆ Es protected through FEC; hence, the weights SNR £ Á ˜ = Á h1 + h2 ˜ , applied at the BS transmitter antennas might be Á 2 2 ˜ N0 Ë ¯ N0 different from the weights the MS expects it to Á h1 + h2 ˜ apply. This causes the composite channel esti- Ë ¯ mate at the MS receiver to be in error. In order where Es/N0 is the symbol SNR based solely on to avoid this situation, verification of the weights transmitted signal energy. In comparison, from is necessary at the MS. Using the channel esti- IEEE Communications Magazine • April 2002 73
  7. 7. eled as an auto-regressive (AR) process [14]. –10 Linear prediction techniques can be used to esti- No diversity mate the AR coefficients and also to predict the –12 OTD future state of the channel. The mobile can cal- Average Ec/lor for 1% FER (dB) STS TXAA culate the feedback based on the predicted –14 future channel state, thus reducing the effect of –16 feedback delay. SCHEMES FOR MORE THAN TWO ANTENNAS –18 The same principles discussed so far for two –20 antenna elements can be used for extensions of closed loop schemes to more than two transmit –22 antennas. One method being contemplated is the direct extension of the filtered phase feed- –24 back scheme in a previous section with a lower feedback rate per antenna. –26 There is a question about the feasibility of 1 10 100 placing many antennas spaced far enough to Velocity (km/h) provide independent fading paths due to space constraints. Closer spacing can induce partial I Figure 6. Performance of TD methods. correlation between diversity paths. A method called the eigen-beamformer has been proposed by Siemens to take advantage of the quasi-sta- Number of base station 2 tionary property of this correlation. The eigen- antennas vectors of the correlation matrix are fed back at a slow rate. The short-term feedback indicates to Carrier frequency 2 GHz the BS some linear combination of the vectors to be used as the antenna weights. A similar con- Bit bate 9600 b/s cept involving multiple banks of beamforming Chip rate 1.2288 Mchip/s antenna arrays has been proposed by Fujitsu. Details of the schemes briefly described in this Walsh code length 128 chips subsection may be found in [15] and its associat- ed references. Convolutional code Rate 1/4, K = 9 Frame duration 20 ms. A COMPARISON OF Pilot Ec/Ior –7 dB TRANSMIT DIVERSITY METHODS Power control On This section compares the performance of dif- Channel estimation Windowed (nonideal) ferent OL and CL methods. The results were generated in a symbol-level simulation environ- Channel model Flat Rayleigh fading ment for the CDMA2000 standard. The simula- tion parameters are given in Table 1. Figure 6 Fading correlation 0 shows the average power per chip required to Feedback error rate 4% transmit at a given frame error rate with power control. It can be seen that the open loop meth- I Table 1. Simulation parameters. ods are robust at higher velocities, while TXAA provides the biggest benefit at the lower veloci- ties. To optimize the system performance the mates from the common pilots as well as the curves in Fig. 6 suggest that a mixture of open dedicated pilot symbols embedded in the traffic and closed loop diversity could be entertained channels, the applied weight may be estimated to combat fast and slow fading, but this would using hypothesis testing. require Doppler estimation at either the BS or Another solution proposed for the feedback MS as well as additional signaling overhead to error problem was to use a decision-directed facilitate dynamic switching between open and method wherein, in case of a frame error, the closed loop TD. erroneous output bits are used to create a repli- ca of the frame and compared with the received frame in order to determine the weights used in CONCLUSIONS each slot in the frame [11]. An attempt has been made to capture the essen- tial elements of transmit diversity in 3G CDMA Feedback Delay — The MS using channel state systems as they are evolving. An overview of the information available to it at any given instant various transmit diversity methods is provided. estimates the required feedack. But there is a Performance comparisons are given, and issues definite delay involved in transmitting the infor- related to these methods were discussed. mation back to the BS. In fast fading conditions, More recently, MIMO technology, which is this delay causes the transmit weights to be out- the use of multiple antennas at both the trans- dated by the time they are applied at the BS mitter and the receiver, is being considered. [13]. One possible solution to this problem is to Polarization diversity, space-time trellis coding use the fact that the fading channel can be mod- and modulation, and the combination of intelli- 74 IEEE Communications Magazine • April 2002
  8. 8. gent beamforming with transmit diversity are BIOGRAPHIES some other technology areas that are promising R. THOMAS DERRYBERRY ( earned Polarization for future evolution. his B.S. (1985) and M.S. (1987) in electrical engineering from the University of Arkansas and his Ph.D. (1995) in diversity, space- REFERENCES electrical engineering from Southern Methodist University. From 1988 to 1998 he held positions with Texas Instru- time trellis coding [1] W. C. Jakes, Microwave Mobile Communications, New ments and Raytheon Systems. In 1998 he joined Nokia York: IEEE Press, 1974. Research Center, Dallas, Texas, where he is currently an and modulation, [2] TIA/EIA IS-2000 Physical Layer Specification for CDMA assistant research manager. He served as the chair of the Spread Spectrum Communications System, June 2000. Adaptive Antennas Ad Hoc group within the 3GPP2, and and the [3] S. M. Alamouti, “A Simple Transmit Diversity Technique remains active in 3GPP2. He is a member of Eta Kappa Nu. for Wireless Communications,” IEEE JSAC, vol. 16, Oct. combination of 1998, pp. 1451–58. STEVEN GRAY earned his B.S. with high honors (1985) and [4] G. J. Foschini, and M. J. Gans, “On Limits of Wireless M.S. (1986) in electrical engineering from Texas A&M Uni- intelligent Communications in a Fading Environment When Using versity, and his Ph.D. (1995) in electrical engineering from Multiple Antennas,” Wireless Pers. Commun., Mar. Northeastern University. From 1986 to 1996 he held posi- beamforming 1998, pp. 311–35. tions with Sandia National Laboratories, E-Systems, and [5] E. I. Telatar, “Capacity of Multi-Antenna Gaussian Chan- The MITRE Corporation. In 1996 he joined Nokia to devel- with transmit nels,” AT&T Bell Labs. tech. rep., June 1995. op CDMA and broadband wireless systems. Currently, he is [6] V. Tarokh, H. Jafarkhani, and A. R. Calderbank, “Space- head of the Radio Communications Laboratory within diversity are Time Block Codes from Orthogonal Designs,” IEEE Nokia Research Center. He is a member of Eta Kappa Nu Trans. Info. Theory, vol. 45, July 1999, pp. 1456–67. and Tau Beta Pi. technology areas [7] S. Bäro, G. Bauch, and A. Hansmann, “Improved Codes for Space-time Trellis-coded Modulation,” IEEE Com- D. MIHAI IONESCU received his M.S. in E.E. (1986) from the that are mun. Lett., vol. 4, Jan. 2000, pp. 20–2. Technical University of Iasi, Romania, and his Ph.D. in E.E. [8] M. Harrison and K. Kuchi, “Open and Closed Loop (1996) from the University of Colorado. After working for promising for Transmit Diversity at High Data Rates on 2 and 4 Ele- the Omnipoint Corporation in Colorado Springs, he joined ments,” 3GPP2-C30-19990817-017, Portland, OR, 1999. Nokia Research Center in Irving, Texas, in 1998, where he future evolution. [9] B. Raghothaman et al., “Performance of Simple Space Time has been conducting research in the area of modulation Block Codes for More than Two Antennas,” Proc. Allerton and coding. Currently, he serves as program chair for the Conf. Commun., Control and Comp., Oct. 2000. IEEE Telecommunications Chapter in Fort Worth, Texas. [10] D. Gerlach and A. Paulraj, “Adaptive Transmitting Antenna Arrays with Feedback,” IEEE Sig. Proc. Lett., GIRIDHAR MANDYAM received his B.S.E.E. degree (Magna Cum vol. 1, no. 10, Oct. 1994, pp. 150–52. Laude) from Southern Methodist University in 1989, his [11] B. Raghothaman, R. T. Derryberry, and G. Mandyam, M.S.E.E. degree from the University of Southern California in “Transmit Adaptive Array without User-Specific Pilot for 1993, and his Ph.D. in electrical engineering from the Uni- 3G CDMA,” Proc. ICASSP 2000, Istanbul, Turkey. versity of New Mexico in 1996. From 1989 to 1998 he held [12] TS 25.214 3GPP TSG RAN WG4, v. 3.2.0, Physical Layer positions with Rockwell International, University of Southern Procedures (FDD), 2000-03. California, Qualcomm International, and Texas Instruments. [13] B. Raghothaman, G. Mandyam, and R.T. Derryberry, “Per- In 1998 he joined Nokia Research Center (Dallas, Texas) formance of Closed Loop Transmit Diversity with Feedback where he is currently a research manager. Delay,” Proc. Asilomar Conf. Sig., Sys. Comp., 2000. [14] T. Eyceoz, A. Duel-Hallen, and H. Hallen, “Determinis- BALAJI RAGHOTHAMAN completed his Bachelor’s degree in elec- tic Channel Modeling and Long Range Prediction of tronics and communication engineering (1994) at Coimbat- Fast Fading Mobile Radio Channels,” IEEE Commun. ore Institute of Technology, India, and received his M.S. Lett., vol. 29, Sept. 1998, pp. 254–56. (1997) and Ph.D. (1999) in electrical engineering from the [15] “Enhancing the Beamforming Feature of the Multiple University of Texas at Dallas. He joined Nokia Research Center Antenna Tx Diversity,” TSGR1 #15(00)-1065, Fujitsu in 1999. He is currently the chair of the IEEE Signal Process- cont. to 3GPP. ing Chapter, Dallas, Texas. He is also a member of Sigma Xi. IEEE Communications Magazine • April 2002 75