Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

IEEE TRANSACTIONS ON COMMUNICATIONS, 2001 (TO APPEAR)

Double Differential Space-Time Block Coding For Time-Selective Fading Channels


Zhiqiang Liu, Georgios B. Giannakis, Fellow, IEEE, and Brian L. Hughes, Member, IEEE

Abstract Most existing space-time coding schemes assume timeinvariant fading channels and offer antenna diversity gains relying on accurate channel estimates at the receiver. Recent single differential spacetime block coding schemes forego channel estimation but are less effective in rapidly fading environments. Based on a diagonal unitary matrix group, a novel double differential space-time block coding approach is derived in this paper for time-selective fading channels. Without estimating the channels at the receiver, information symbols are recovered with antenna diversity gains regardless of frequency offsets. The resulting transceiver has very low complexity and is applicable to an arbitrary number of transmit and receive antennas. Approximately optimal space-time codes are also designed to minimize bit error rate. System performance is evaluated both analytically and with simulations. Index TermsDiversity methods, transmit antennas, space-time codes, group codes, differential encoding

I. I NTRODUCTION The rapidly growing demand for reliable high data rate transmissions over fading channels has stimulated much interest in multiple-antenna communications. Particularly popular recently are space-time coding schemes that rely on multiple-antenna transmissions and appropriate signal processing at the receiver to provide diversity gain and signicantly increased channel capacity [2, 18, 19]. So far, the effectiveness of most space-time coding schemes relies on accurate channel estimates at the receiver, which are either acquired through training sessions [11], or, by employing blind estimation algorithms [12]. Since multiple channels are involved with multiple-antenna links, channel estimation implicitly assumes that the underlying channels remain invariant for sufciently long time - an assumption that may not be satised in rapidly fading mobile environments. Single differential space time (SDST) block coding schemes were proposed in [7, 8, 20] to achieve diversity gains without channel state information (CSI). By obviating channel estimation, SDST coding schemes allow for slowly changing channels that have to remain invariant within two consecutive blocks. Their performance however, degrades when channel frequency offsets (errors) are present. In wireless mobile communications, frequency offset arises due to relative motion between the transmitter and the receiver
Manuscript received March, 2000. This work was supported by ARO grant no. DAAG55-98-1-0336. This paper was presented in part at Wireless Communications and Networking Conference, Chicago, IL, September 2000. Z. Liu and G. B. Giannakis are with the Department of Electrical and Computer Engineering, University of Minnesota, 200 Union Street SE, Minneapolis, MN 55455. Tel/fax: (612)626-7781/625-4583; Emails: [lzq,georgios]@ece.umn.edu. B. L. Hughes is with the Department of Electrical and Computer Engineering, North Carolina State Univ., Raleigh, NC 27695-7914; Email: blhughes@eos.ncsu.edu; Work of this author was supported in part by the NSF under grants CCR-9903107 and CCR-9805117, and by the Center for Advanced Computing and Communication. Publisher Item Identier

and is proportional to the Doppler shift. For example, it is anticipated that the third-generation European cellular standards km/hr [3]. If the will operate on trains moving as fast as carrier frequency is GHz, the induced Doppler shift may be Hz. Another source of frequency offset is the carrier up to frequency mismatch between transmit- and receive-oscillators that arises due to drifts from their nominal frequency. Traditionally, one technique to mitigate frequency offset is to estimate it at the receiver using a frequency acquisition and tracking circuit [17] or sophisticated harmonic retrieval digital algorithms [4, 13]. Clearly, the success of such techniques depends on estimation accuracy, which is ultimately limited by the rate at which the frequency offset changes, the circuits complexity and the cost which the receiver can afford. In order to maintain receiver simplicity especially at the mobile where power consumption and size/cost are limited, an alternative technique to overcome frequency offsets is the single-antenna double differential coding which relies on proper encoding at the transmitter that renders decoding at the receiver insensitive to these offsets [5, 14, 16, 17]. Unfortunately, similar to coherent single-antenna communications, the single-antenna double differential coding can not handle effectively frequency offsets in the presence of fading. The latter motivates our double differential space-time coding herein.

Generalizing single-antenna double differential coding ideas to our space-time context, we develop a novel double differential space time (DDST) coding scheme along with pertinent modulator design criteria and error bounds. Our model entails time-selective channels which are allowed to vary as fast as one symbol duration. Based on diagonal unitary group code matrices, properly designed encoding at the transmitter enables information symbol recovery at the receiver with diversity gains and without channel knowledge. The resulting algorithm has low complexity and can be applied to an arbitrary number of transmit and receive antennas. We also design an optimal space-time code matrix that approximately minimizes bit error rate (BER). Interestingly, the optimal DDST codes coincide with those proposed in [8] for SDST coding. In addition to simplicity, our scheme enables dB performance improvement when estimates of frequency offsets are available, while dB performance gain is possible with perfect CSI at the receiver. Thorough simulations illustrate robust performance of our design even in the presence of time-selectivity.

The rest of this paper is organized as follows. In Section II we describe the time-selective channel and derive our data model. The double differential space time coding/decoding is developed in Section III. In Section IV, the approximately optimal code matrix designs are derived by minimizing the bit error rate and

IEEE TRANSACTIONS ON COMMUNICATIONS, 2001 (TO APPEAR)


ST Encoder


Receive Antennas

Transmit Antennas

Gaussian noise (AWGN) with two-sided power spectral density ; and is the transmit amplitude. We assume that the channel delay spread is small compared to but the channel coherence time is comparable to . Under as: these assumptions, we can model the channel

(3)

Fig. 1. Baseband system model

some examples are given. Supporting simulations are presented in Section V, and Section VI concludes this paper. II. S YSTEM M ODELING Fig. 1 depicts a wireless communication system equipped with transmit antennas and receive antennas. At the transmitter, each information symbol belonging to a nite alphabet set is encoded by the space-time encoder which uniquely maps to the following code matrix:


. . .


.. . . . .

(1)

where: is the Doppler frequency shift caused by relative motion between the transmitter and the receiver; and captures multipath fading effects induced by the channel from the th transmit antenna to the th receive antenna. It is assumed that remains invariant during at least three consecutive information symbols. To obtain (3), we also assume that the Doppler frequency (being independent of ) is common to all transmit antennas. This assumption is valid only if the multipath components originate far away from the receiver so that they all share a common angle of arrival. The effects of s will be revisited having transmit antennas with different in the simulations Section V. Under the channel model (3), the received signal (2) can be written as:

The rows of are generated in successive time intervals of duration . Each of the elements in a given row is forwarded to one of the transmit antennas. At each time slot , the th transmit antenna pulse-shapes the coded symbol using the transmit-lter and then modulates it onto the carrier frequency. The modulated signals from all transmit antennas are sent simultaneously with the same transmission period . We emphasize that encoded symbols from each antenna are transmitted for every information symbol. Thus, the overall transmission rate is bits/sec/Hz, where denotes the cardinality of . denote the unknown time-varying channel from Let the th transmit antenna to the th receive antenna. The transmitted signal from the th transmit antenna propagates through , and at the th receive antenna, it is ltered by the receive-lter that is matched to . We dene to be the overall impulse response of the time-varying channel including the transmit and receive lters; i.e., with denoting convolution. We as sume perfect symbol timing but not carrier frequency and phase synchronization at the receiver. At the th receive antenna, the baseband received signal from all transmit antennas is therefore given by:


(4)

. Similar to single-antenna links, (4) where is valid so long as the effect of on the receive lter can be neglected, i.e., is small as compared to the bandwidth of . If the received signal is sampled at the symbol rate , the discrete-time received sequence is given by:

(5) whereand sothat and satises the Nyquist-. By choos condition, ing we have where denotes Kroneckers delta. In this case, (5) reduces to:

(6)


(2)

where: and denote the carrier frequency and phase offsets introduced by the imperfect knowledge of the carrier at the where is the additive white receiver;

where: the noise becomes white complex Gaussian with variance per dimension; the normalized frequency error ; and in (5) is absorbed into . We will nd it convenient to dene the received data block of length , and cast (6) into a matrix/vector form as follows:

(7)

IEEE TRANSACTIONS ON COMMUNICATIONS, 2001 (TO APPEAR)

and
where diag

Data model (7) incorporates the Doppler effects, the carrier frequency offsets through s and channel fading effects through s. If one can estimate s accurately at the receiver, the frequency errors can be compensated for and then the information symbols can be recovered with or without knowledge of s as discussed in [8]. Estimating s is a challenging task and has been addressed by many authors in a single transmitsingle receive-antenna setting (see e.g., [13] and [4]). However, targeting non-coherent reception, our goal herein is to design space-time encoders and decoders so that the information data can be retrieved with diversity gains at the receiver regardless of the channel variations modeled by (7). III. D OUBLE D IFFERENTIAL ST C ODING

, .

In the second recursion, the matrix conveys our information symbols and is chosen to correspond one-to-one at the receiver, with . Hence, knowing or detecting determines uniquely . This implies that we should deand to be one-to-one and as sign the mapping between a rst step in this direction, we choose equal cardinalities: . Next, we dene and , and establish, without loss of generality, the following one-to-one (ordered) mapping between elements of and :

, we design in this section Given the received data block to recover the information symbols withthe code matrix out requiring knowledge of s and s. Motivated by the conventional single-antenna double differential coding approaches of [5, 14, 16, 17], where three consecutive samples are considered to remove the unknown frequency and phase errors, our basic idea is to process three consecutive received data blocks, , and from each receive annamely, tenna, to recover the data symbol by exploiting a judiciously , and designed encoding relationship among . As discussed in [10], for high signal-to-noise ratio (SNR), the code matrix (1) with orthogonal columns of equal-energy approaches channel capacity in multiple-antenna communication links through at fading channels. Accordingly, for our timeselective setup we will focus on code matrices that satisfy:

AND

D ECODING

and the fact that Due to the group structure of , we deduce from (10) that: . Thus, (9) is the same as the recursion in [8, Eq. (12)], which enables single differential decoding of the information data when the underlying channels exhibit at fading. Note also that choosing to satisfy (8), implies (by induction) that generated by (9) will satisfy (8) as well. Using our space-time encoder generated by (9) and (10), three consecutive received blocks of samples at the th receive antenna can be written as [c.f. (7), (9), (10)]:

(11)

(12a)

(12b)

identity matrix. Checking the dimensionality of , . Since we infer that for (8) to hold we have to select , we choose in order to maximize

where

stands for Hermitian transpose and

(12c)

(8) denotes the

where in establishing (12b) and (12c) we exploited the fact that , and are all diagonal matrices so that they commute.

A. Maximum Likelihood Detection

the transmission rate. For time-invariant channels, two consecutive code matrices in the SDST modulation of [7, 8] are related through a unitary matrix group. The group structure leads to uniform performance over all code matrices and reduces complexity at the receiver, both of which are also desirable in our design. Similar to [7, 8], our approach to double differential space-time (DDST) coding is also based on a unitary matrix group. Unlike [7, 8] however, we will nd it necessary to restrict our unitary matrices to be diagonal. Let be any nite group of , unitary and diagonal matrices ( ). We design our space-time code matrices to satisfy the following recursion: (9)

Based on the received data blocks , and from each receive antenna, we derive next our without knowledge of s and detector to decode s. We start by dening the vector , the augmented re ceived data vector at the th receive antenna and the corresponding matrix . Using these denitions and (12a)-(12c), we can write:

where the so-termed generating matrix sion:

obeys also a recur

where is dened similar to . So far, we focused on the th receive antenna only. To allow for multiple receive , antennas, we further dene , and the block matrix:


.. .

(13)


. . .

(10a) (10b)

. . .

(14)

IEEE TRANSACTIONS ON COMMUNICATIONS, 2001 (TO APPEAR)

Based on these denitions, the counterpart of (13) for receive antennas is:

(15)

In order to gain insight of good decoder structures, we rst conconsider the probability density function (pdf) of and . We assume that s ditioned on are i.i.d., zero-mean, unit-variance complex Gaussian variables and are independent from s. Since the matrix is unitary and the pdf of depends only on , we infer that and have the same pdf. Because and are independent complex Gaussian is given by [8, 10]: random vectors, the conditional pdf of

Let us start by relating the blocks , and th receive antenna. It follows from (12a)-(12c) that:

B. Suboptimum Double Differential Space Time Decoding

from the

(18)

on

det (16) where from (15) the covariance matrix and det stands for matrix determinant. Noting that and using the matrix identity det det , we nd that the det does not depend on and . Further, ap plying the matrix inversion lemma , we express: (19) We also observe that the conditional pdf depends Since and are both diagonal matrices, we null the off-diagonal elements on both sides of (19) to obtain:
1

and have been reObserve from (18) that the unknowns moved using the single differential space-time coding design in (9). This step is similar to the single differential space-time modulation of [8]. However, the unknown frequency error is still present in (18). If one has accurate estimates of , the information symbols can be retrieved using the detection scheme in [8]. Looking for a non-coherent space-time coding/decoding scheme, we cancel from (18) by performing the outer-product:

only through the quadratic form:

diag

ReTr

where OT denotes other terms which do not depend on and , and ReTr stands for the real part of the trace operator. Equation (17) provides important insight into the detectors and are diagonal matrices, the design: First, since , and optimal detector will depend on only through the diagonal elements of their outer products; Secs, ond, because there are no cross-terms among different data processing for different receive antennas can be separated. be the matrix obtained by nulling Let diag the off-diagonal elements of the square matrix , where is the Schur (or Hadamard) product. It follows from (17) , that for the th receive antenna diag and diag will be suf diag cient statistics for the detection of and . However, since the unknown s are still present in (17), the maximumlikelihood (ML) detector based on (16) is not feasible. For this reason, we focus on a sub-optimal double differential decoding and scheme based on the statistics: diag , for diag .

OT

diag


(20)

(17) diag

where in deriving the second equality disappears since: . For brevity, we collect all noise terms in (20) to the right hand side and arrive at:


diag

(21) where the composite noise matrix is given by: diag diag
diag

(22)

, ,

For

matrices of appropriate dimensions, it holds that .

Eq. (21) conrms that for the th receive antenna the detection of depends only on the statistics: diag . For notational simplicity, we let and diag denote a vector obtained from the diagonal elements of the matrix , and rewrite (21) as:

(23)

IEEE TRANSACTIONS ON COMMUNICATIONS, 2001 (TO APPEAR)

diag and . Clearly, (23) is equivalent to (21) when it comes to detecting from and . Since (23) is satised for all receive-antennas, we concatenate s for and write for all receive antennas the supervector: where

where:

(24) ,
and

is AWGN and we assumed it to be independent for different s and s, it can be veried by direct substitution from (22) that is colored. Thus, the ML detection of from (24) has to involve multiple blocks which leads to exponential
complexity in the number and size of the blocks involved. Motivated by the conventional double differential coding for single-antenna transmissions [17], we will develop a subfrom optimal yet low-complexity detector, which recovers (24) using three successive received blocks from each receive antenna. To explain how it works at high SNR, we neglect the and approximate by a second-order noise terms in zero-mean complex Gaussian vector whose covariance matrix can be obtained from (22), (23) as:

stands for Knonecker product.

Although

Substituting (21) into (29), it can be easily veried that the detection rule in (29) recovers uniquely in the absence of noise. Noise effects will be investigated by simulations in Section V. The DDST decoder based on (29) processes three consecutive received data blocks and has very low complexity. To gain further insight about our DDST detector in (29), we consider next two special cases: Case 1: In a single transmit- and receive-antenna setting ( ), the matrix reduces to a scalar which we denote by . Thus, the detector in (29) can be rewritten as:

(30) Note that the unitary group property of implies that must be a PSK symbol. We omit the denominator in (30) which does not affect our detection of from (30), and observe that

where diag , for . Since is complex Gaussian, it follows from (24) that the pdf of conditioned on and is given by: det (26) Given and , our DDST detector chooses to
maximize the conditional pdf in (26), i.e.,

.  . .


.. .

. . .

(25)

(30) reduces to a conventional single-antenna double differential detector [5, 17]. Hence, the conventional single-antenna double differential coding is a special case of our scheme. Case 2: Observe from (29) that the performance of our detector does not depend on as long as it satises (8). As special a to be diagonal, e.g., . case, we choose is diagonal, it follows from (9) that is also diagSince onal . Thus, our system described in Section II reduces to an antenna-switching scheme where only one of the transmitantennas is active each time. The advantage of using a diagois to allow different transmit-antennas to have possibly nal different Doppler shifts. In this case, we can model the chan nel as: , where is the Doppler frequency shift from the th transmit-antenna to s are diagonal, followthe th receive-antenna. When the ing steps similar to (4), (5) and (6), we arrive at the following counterpart of (7):

(27)

The conditional likelihood maximization in (27) for the pdf in (26) reduces to a minimum distance or a correlation maximization if one omits the terms that do not depend on . The resulting detector is given by:

Re ReTr and after substituting for from (24), we end up with


diag diag

(28)

ReTr diag


(29)

. With minor modications, it can be tary diagonal matrix shown that the steps used to derive (29) from (7) apply also to is (31). In other words, our DDST detector with diagonal applicable even when different transmit antennas have different Doppler shifts. However, with diagonal code matrices, only a single antenna transmits every time. If we maintain the same total transmit power for both diagonal and non-diagonal code matrices, the scheme with diagonal code matrices will have higher peak power, which is undesirable in some applications. The depends on how different the Doppler frequency choice of shifts are for the different transmit antennas. If the differences should be preferred. Othare negligible, the non-diagonal erwise, the diagonal is recommended. In the sequel, we will focus on model (7). Modications to accommodate model (31) are straightforward and will not be detailed.

(31) diag and where . Compared with (7), the diag is replaced by the unidifference in (31) is that

IEEE TRANSACTIONS ON COMMUNICATIONS, 2001 (TO APPEAR)

We can now summarize our DDST coding and decoding stages in the following steps: s1) Choose and design to satisfy ; s2) Build a unique mapping between and as in (11); s3) Choose to satisfy (8) and ; and ; s4) Transmit code matrices , , using the mapping in s2); s5) Map to using (9), (10), and transmit ; s6) Obtain s7) Detect via (29) and recover by mapping back to as in (11).

For notational simplicity, we let

and re-write (35) as: Re (36)

and

is a zero-mean, complex where it can be easily shown that ; thus, Gaussian random vector with covariance matrix Re is a zero-mean Gaussian variable with vari . We can now compute ance as:

So far, we required to be unitary and diagonal in order to arrive at the DDST coding/decoding scheme we described in steps s1)-s7). We proceed next to design the group aiming at optimal performance in terms of minimizing the BER. IV. D IAGONAL U NITARY G ROUP D ESIGNS As discussed in Section III, the information symbols are conveyed through the code matrix whose group structure leads to the low-complexity DDST decoder in (29). In addition to low-complexity, well designed code matrices help us achieve (or approximate) the optimal performance affordable by the system design. Because BER minimization is our ultimate criterion, we motivate our group design criterion by analyzing the error performance of our detector in (29). Dropping the block index , we dene the pairwise code ma as the event that the receiver decodes trix error event when is actually sent. The BER given that code matrix is transmitted is union bounded by [6]:

(37)

where denotes the Q-function. To obtain a tractable expression of as a function of , we will further invoke the high SNR assumption to drop . Specically, writing, noise terms from diag

Tr

number of error bits in event number of information bits per code matrix

(32) is given

diag we can approximate as:


Tr diag

(38)

where the pairwise code matrix error rate by [c.f. (26)]:

(39)

(33)

, and where . Substituting (39) into denotes the th diagonal entry of is Chernoff bounded by: (37),

Similar to [8, 18], we note that at high SNR only adjacent code matrix errors occur. Here, we use this fact to approximate the BER in (32) by:

(40)

is the adjacent pairwise code matrix error where in (32); and is a constant rate that dominates that accounts for the ratios in (32) and the number of code matrices adjacent to . Since the BER is mainly determined by , we choose as our gure of merit and proceed to design our diagonal unitary group that minimizes . can be expressed as: Using (24), (26) and (33),

(34)

, , are i.i.d, Recall now that zero-mean, unit-variance complex Gaussian variables. Because is unitary and the pdf of depends only on , we infer that the s are also i.i.d, zero-mean, unitvariance complex Gaussian variables. We average (40) with respect to s, and compute the average adjacent pairwise code matrix error probability as:

Re

(35)

det

(41)

IEEE TRANSACTIONS ON COMMUNICATIONS, 2001 (TO APPEAR)

before the SNR, the bound in (41) is Except for a factor identical to that in [8, Eq. (18)]. This suggests that our DDST decoder suffers a dB performance loss when compared to [8]. As mentioned in Section II, the constant is used to control the , where transmit power. Suppose our alphabet size is is a positive integer and let denote the bit-energy. We then and (41) can be rewritten as: have

2 4 8 16
O PTIMAL

0.5 1 1.5 2

, , ,
TABLE I

0.5 0.25 0.179 0.0732

det

DIAGONAL UNITARY GROUP (

(42)

1/3 2/3 1 4/3


O PTIMAL

The expression in (42) has important implications for the design takes of our diagonal unitary group. At high SNR, , where and depend the form on the difference . The so-termed transmit diversity advantage is dened as:

2 4 8 16

, , , , , , ,

0.333 0.167 0.0879 0.067

Rank

(43)

DIAGONAL UNITARY GROUP (

TABLE II

and measures the performance gain from multiple transmissions. Since and are both diagonal matrices, in order to , we should design achieve maximum for . With , the so-termed coding is given by advantage

det

, every full-rank group of Lemma 1 [9]: For diagonal unitary matrices with is equivalent to an cyclic group, for some odd numbers . In particular, letting

(44)

In particular, when , our DDST decoder reduces to the conventional single antenna double differential coding scheme. , then . Compared to the corIn this case, if responding coherent decoding scheme [15, p. 775], our scheme shows a dB performance loss. In order to optimize our DDST performance, we should max. Correspondingly, we will pursue the imize both and following code design criteria: , c1) Design the group such that: , . det c2) Design the group such that: is maximized. Note that the criteria c1) and c2) are identical to those used in designing the group codes of [8]. Our main difference here is that we have to restrict our group code matrices to be diagonal. This implies that if a diagonal unitary group is optimal for the DDST coding, then it is also optimal for the SDST coding , it is proved in [8, Lemma scheme. For and must form a A4] that all diagonal groups with cyclic group, for some odd . Specically, letting and the matrix

.  . .

. . .

..


. . .

(47)

the diagonal unitary group can be represented as:


and

(48)

Using Lemma 1, it is easy to verify that

(49)

Thus, designing our diagonal unitary group codes is equivalent to choosing is maximized. We collect such that . in Tables I and II the optimal codes for Having designed the group , we next resort to simulations in order to test the performance of the DDST coding/decoding and compare it with existing alternatives.

V. S IMULATIONS We simulate the performance of our DDST coding scheme using BER as our gure of merit which we average over channel and noise realizations for each point. The timeselective fading channels are generated by taking as comfor both real and plex Gaussian variables with variance as a random deviate uniformly disimaginary parts, and . The initial code matrix is chosen tributed between and as:

(45)

the unitary diagonal group can be specied as:

(46)

This result has been generalized recently to an arbitrary number of transmit antennas as we quote in the following lemma:

IEEE TRANSACTIONS ON COMMUNICATIONS, 2001 (TO APPEAR)

10

SNR=16 dB, R=1.0 bit/sec/Hz

10

ddst single-antenna

10
10
-1

-1

BER

BER
10
-2

10

-2

10

-3

10

-3

0.05

0.1 0.15 Frequency Error (fk )

0.2

0.25

10

-4

10 E /N (dB)
b 0

12

14

16

18

20

Fig. 2. Single differential space-time codes with frequency errors

(a)

R=2

Example 1 (performance of single differential space-time modulation [8] with frequency errors): To motivate the design of double over single differential space-time coding, we investigate the performance of SDST coding when the frequency error does exist, but is not compensated for. In this example, we consider a system with transmit antennas and a single dB. The optimal group receive antenna at SNR (see Table I) are employed. The simulation codes with results are shown in Fig. 2 where we observe that the SDST coding is very sensitive to frequency errors. When , the , the system BER increases from to . When . For voice transmissions, the data is useless because BER ). Thus, kbps ( is equivrate is about alent to Hz which may happen in practice especially when the carrier frequency is up to GHz. Example 2 (performance comparison with single-antenna double differential codes [5]): We test our system with transmit antennas and a single receive antenna, and compare it with single-antenna double differential coding. We choose QPSK for our system and BPSK for the single-antenna double differential coding so that for both systems. Fig. 3(a) shows that the DDST coding outperforms the singleantenna double differential coding considerably especially at high SNR values. Similarly, we choose QPSK for the singleantenna double differential coding system and PSK for our system. Both systems have and Fig. 3(b) is the counterpart of Fig. 3(a) for this rate. Since we employ different modulations for a single antenna and two antennas, the performance gain of QPSK over PSK is bigger than that obtained from transmit diversity which is small at low SNR. Example 3 (performance comparison with different unitary codes): The unitary diagonal group codes are designed to maxin (44). In order to verify our design criteria and imize investigate the role optimal codes play in our scheme, we compare the DDST schemes with different group codes. The opticodes will be compared to codes. Two mal transmit-antennas and a single receive-antenna are used for both

10

ddst single-antenna

10

-1

BER

10

-2

10

-3

10

-4

10

15 E /N (dB)
b 0

20

25

30

(b)

Fig. 3. Space-time double differential codes vs. single-antenna double differential codes

codes. Checking the coding advantage of the (16;1,1) codes, which is approximately of that we nd for the optimal codes. This implies that in theory codes may induce about dB loss compared to the codes. The simulation results in Fig. 4 show that the cyclic group our optimal group codes outperform the codes by about dB. Considering the approximations we have made to arrive at (41), the simulation results are close to our theoretical results. Example 4 (performance comparison with different number of receive antennas): We use the same parameters as in Exam) with two antenna ple 2 and compare a single antenna ( ( ) reception. The results are depicted in Fig. 5 where a is observed, signicant performance improvement for as expected.

IEEE TRANSACTIONS ON COMMUNICATIONS, 2001 (TO APPEAR)

10

R=2

(16; 1, 7) (16; 1, 1)

10

-1

10

-2

10

-3

10

-4

10

15 Eb/N0 (dB)

20

25

Fig. 4.

versus Group Codes in DDST


R=1

10

Example 5 (performance when is different for different transmit antennas): In deriving (7), we assumed that the Doppler frequency shift is common to all transmit anten nas. To test robustness against cases where s are different for different transmit antennas, we simulate our system with transmit antennas and a single receive antenna. The optimal group codes are used. Let denote the difference between s for the two transmit antennas. We emphasize that our system is designed for a common ; thus, should be taken into account. We check perforonly the mance when and when . Fig. 6 shows that increases the BER from to at dB. led the single differential space-time codRecall that ing scheme in Example 1 to an unacceptable performance loss. To investigate the effects of different s when , by the diagonal matrix and repeat the we replace . Fig. 6 conrms that there is no persimulation with being diagonal when . The formance loss with depends on the intended application, as we dischoice of cussed in Section III.

BER

N =1 r N =2
r
1

VI. C ONCLUSIONS Based on a diagonal unitary matrix group, we derived a lowcomplexity double differential space-time coding scheme suitable for time-selective channels and applicable to an arbitrary number of transmit and receive antennas. Without channel state information available, double differential space-time coding was shown capable of providing diversity gain in the presence of unknown frequency offsets. Optimal space-time codes were designed to minimize approximately the BER. Single and double differential space-time coding for doubly selective channels constitutes a natural generalization of this work and relevant results will be reported elsewhere.
20

10

BER

10

10

10

10 Eb/N0 (dB)

12

14

16

18

R EFERENCES
S. Benedetto and E. Biglieri, Principles of Digital Transmission, Kluwer Academic/Plenum Publishers, 1999. [2] G. J. Foschini and M. J. Gans, On limits of wireless communications in a fading environment when using multiple antennas, Wireless Personal Communications, vol.6, no.3, pp.311-335, March 1998. [3] FRAMES Muliple access proposal for UMTS radio interface-SMG2, Workshop on UMTS Radio Interface Technologies, December 1996. [4] F. Gini and G. B. Giannakis, Frequency offset and symbol timing recovery in at-fading channels: a cyclostationary approach, IEEE Trans. on Communications, vol. 46, no. 3, pp. 400-411, March 1998. [5] F. Gini and G. B. Giannakis, Generalized differential encoding: a nonlinear signal processing perspective, IEEE Trans. on Signal Processing, vol. 46, no. 11, pp. 2967-2974, November 1998. [6] J. C. Guey, M. P. Fitz , M. R. Bell and W. Y. Kuo, Signal design for transmitter diversity wireless communication systems over Rayleigh fading channels, IEEE Trans. on Communications, vol. 47, no. 4, pp. 527537, April 1999. [7] B. Hochwald and W. Sweldens, Differential unitary space-time modulation, IEEE Trans. on Communications, vol. 48, no. 12, pp. 2041-2052, December 2000. [8] B. L. Hughes, Differential space-time modulation, IEEE Trans. on Infomation Theory, vol. 46, no. 7, pp. 2567-2578, November 2000. [9] B. L. Hughes, Optimal space-time constellations from groups, to be submitted to IEEE Trans. on Infomation Theory, 2000. [10] B. M. Hochwald and T. L. Marzetta, Capacity of a mobile multipleantenna communication link in Rayleigh at fading, IEEE Trans. on Information Theory, vol. 45, no. 1, pp. 139-157, January 1999. [11] Y. Li, N. Seshadri, S. Ariyavisitakul, Channel estimation for OFDM systems with transmitter diversity in mobile wireless channels, IEEE Journal [1]

Fig. 5.

versus
R=1

10

10

BER

10

10

f =0 k fk=0.1 diag. C(0) f =0.1 nondiag. C(0)


k

10

10 E /N (dB)
b 0

12

14

16

18

20

Fig. 6. Performance with

IEEE TRANSACTIONS ON COMMUNICATIONS, 2001 (TO APPEAR)

10

[12]

[13] [14] [15] [16] [17] [18]

[19] [20]

on Selected Areas in Communications, vol.17, no.3, pp. 461-471, March 1999. Z. Liu, A. Scaglione, S. Barbarossa, G. B. Giannakis, Transmit-antennae space-time block coding for generalized OFDM in the presence of unknown multipath, IEEE Journal on Selected Areas in Communications, 2001 (to appear); see also Proc. of 33rd Annual Asilomar Conference on Signals, Systems, and Computers, vol. 1, pp. 1557-1561, Pacic Grove, CA, October 1999 U. Mengali and A. N. DAndrea, Synchronization Techniques for Digital Receivers, Plenum, NY, 1997. Y. B. Okunev, Theory of Phase-Difference Modulation, Svyaz Press, Moscow, 1979. J. G. Proakis, Digital Communications, Third Edition, Mc-Graw Hill, 1995. M. Pent, Doubly differential PSK scheme in the presence of Doppler shift, Digital Communications in Avionics, AGARD Proc., pp. 43.1-43.11, 1978. M. K. Simon, S. M. Hinedi and W. C. Lindsey, Digital communication techniques, signal design and detection, Prentice Hall, 1995. V. Tarokh, N. Seshadri and A. R. Calderbank, Space-time codes for high data rate wireless communication: performance criterion and code construction, IEEE Trans. on Information Theory, vol. 44, no. 2, pp. 744-765, March 1998. V. Tarokh, H. Jafarkhani, A. R. Calderbank, Space-time block codes from orthogonal designs, IEEE Trans. on Information Theory, vol.45, no.5, pp. 1456-1467, July 1999. V. Tarokh and H. Jafarkhani, A differential detection scheme for transmit diversity, IEEE Journal on Selected Areas in Communications, vol. 18, no. 7, pp. 1169-1174, July 2000.

and Array Processing Committee. He is a member of the Editorial Board for the Proceedings of the IEEE, he chairs the SP for Communications Technical Committee and serves as the Editor in Chief for the IEEE Signal Processing Letters. He is a Fellow of the IEEE, a member of the IEEE Fellows Election Committee, the IEEE-SP Societys Board of Governors, and a frequent consultant for the telecommunications industry.

Zhiqiang Liu received the B. S. degree from the Department of Radio and Electronics, Peking University, China, in 1991 and the M. E. degree from the Institute of Electronics, Chinese Academy of Science, China, in 1994. From 1995 to 1997, he worked as a research scholar at the Department of Electrical Engineering, National University of Singapore. From 1997 to 1998, he was a research assistant with the University of Virginia. He is currently working toward the Ph. D. degree at the Department of Electrical and Computer Engineering, University of Minnesota, Twin Cities. His research interests include space-time coding, multi-user detection, multi-carrier, and blind channel estimation algorithms.

Brian Hughes (S84-M85) was born in Baltimore, MD, on July 16, 1958. In 1980, he received the B.A. degree in mathematics from the University of Maryland, Baltimore County. He received the M.A. degree in applied mathematics as well as the Ph.D. degree in electrical engineering from the University of Maryland, College Park, in 1983 and 1985, respectively. From 1980 to 1983, he worked as a mathematician at the NASA Goddard Space Fight Center in Greenbelt, MD. From 1983 to 1985, he was a Fellow with the Information Technology Division of the Naval Research Laboratory in Washington, DC. From 1985 to 1997, he was Assistant and then Associate Professor of Electrical and Computer Engineering at The Johns Hopkins University of Baltimore, MD. In 1997, he joined the faculty of North Carolina State University in Raleigh, where he is currently Associate Professor of Electrical and Computer Engineering. His research interests include communication theory, information theory, and communication networks. Dr. Hughes has served as Editor of the IEEE Information Theory Newsletter, Associate Editor for Detection Theory of the IEEE Transactions on Information Theory, and is currently Editor for Theory and Systems of the IEEE Transactions on Communications. He is Co-Chairman of the 1987 and 1995 Conferences on Information Sciences and Systems, and served on the program committees of the 1993 and 1995 IEEE International Symposia on Information Theory, and the 1999 and 2000 IEEE Wireless Communications and Networking Conference.

Georgios B. Giannakis (F97) received his Diploma in Electrical Engineering from the National Technical University of Athens, Greece, 1981. From September 1982 to July 1986 he was with the University of Southern California (USC), where he received his MSc. in Electrical Engineering, 1983, MSc. in Mathematics, 1986, and Ph.D. in Electrical Engineering, 1986. After lecturing for one year at USC, he joined the University of Virginia in 1987, where he became a professor of Electrical Engineering in 1997. Since 1999 he has been with the University of Minnesota as a professor of Electrical and Computer Engineering. His general interests span the areas of communications and signal processing, estimation and detection theory, time-series analysis, and system identication subjects on which he has published more than 120 journal papers, 250 conference papers and two edited books. Current research topics focus on transmitter and receiver diversity techniques for single- and multi-user fading communication channels, redundant precoding and space-time coding for block transmissions, multicarrier, and wide-band wireless communication systems. G. B. Giannakis has received three best paper awards from the IEEE Signal Processing (SP) Society (1992, 1998, 2000) and the Societys Technical Achievement Award in 2000. He co-organized three IEEE-SP Workshops (HOS in 1993, SSAP in 1996 and SPAWC in 1997) and guest (co-) edited four special issues. He has served as an Associate Editor for the IEEE Trans. on Signal Proc. and the IEEE SP Letters, a secretary of the SP Conference Board, a member of the SP Publications Board and a member and vice-chair of the Statistical Signal

You might also like