IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO.

4, APRIL 2008

1227

Adaptive DS-CDMA Receiver with Code Tracking in Phase Unknown Environments
Fang-Biau Ueng, Jun-Da Chen, and Shang-Chun Tsai, Student Member, IEEE
Abstract— In this paper, we propose and analyze a new noncoherent receiver with PN code tracking for direct sequence code division multiple access (DS-CDMA) communication systems in multipath channels. We employ the decision-feedback differential detection method to detect MDPSK signals. An “error signal” is used to update the tap weights and the estimated code delay. Increasing the number of feedback symbols can improve the performance of the proposed noncoherent receiver. For an infinite number of feedback symbols, the optimum weight can be derived analytically, and the performance of the proposed noncoherent receiver approaches to that of the conventional coherent receiver. Simulations show good agreement with the theoretical derivation. Index Terms— DS-CDMA, code tracking, decision-feedback differential detection.

Despreader U(n) Adaptive filter C(n)

^

S

d coh (n)

d coh (n)

-

+

+

PN sequence generator

code delay estimator

PN code tracking loop

Fig. 1. The conventional coherent receiver - joint detection and PN code tracking.

I. I NTRODUCTION Pread-spectrum communication needs to synchronize the spreading waveforms in transmitting and receiving ends. If the two waveforms are out of synchronization by a little chip time, insufficient signal energy will reach the receiver, the performance of data demodulation is thus degraded. Synchronizing the DS-SS system includes a two step procedure, code acquisition and code tracking. Although code acquisition is an important problem, the development of some techniques for accurate code tracking plays an equally important role in supporting the acquisition process once the code has been acquired. In [1]-[4], some adaptive code tracking techniques were proposed. In [1], the authors have proposed the scheme that performed both acquisition and tracking with the same circuitry, therefore a significant simplification in the overall DS-SS receiver structure is gained. In [2], an FIR adaptive filter that mitigates the effect of multipath on delay estimation was utilized. The tap weight vector of this adaptive filter can be used to provide accurate estimation of the multipath delays. Hosemann et al. [3] presented a new tracking scheme for handling multiple access interference (MAI) effect without the requirement of a pilot channel or training symbols. In [4], a coherent timing error detector embedded in a code tracking loop for Rake reception was presented. Its filter coefficients are computed online in order to minimize an interference cost function. However, in these references, multipath and MAI

S

Manuscript received October 8, 2006; revised April 16, 2007; accepted June 7, 2007. The associate editor coordinating the review of this paper and approving it for publication was S. Kishore. This work was supported in part by the National Science Council of Taiwan under Contract NSC96-2221-E005-004. The authors are with the Department of Electrical Engineering, National Chung-Hsing University, 250, Kuo-Kuang Road, Taichung, Taiwan (R.O.C.) (e-mail: fbueng@nchu.edu.tw; {d9064102, d9264104}@mail.nchu.edu.tw). Digital Object Identifier 10.1109/TWC.2008.060803.

effects are not considered simultaneously. This paper deals with tracking issue and assumes successful initial acquisition before tracking process. Therefore, the received signal has only a small timing offset (±1/2 chip duration) from the correct timing. In this paper, we refer to the adaptive Pseudo-Noise (PN) code tracking scheme proposed in [5][6] with well-known coherent detection as conventional coherent (closed-loop time-delay estimation) receiver. The conventional coherent receiver deals with data detection and PN code timing recovery jointly is described in Fig. 1. PN code tracking can be categorized into coherent and noncoherent loops. When the demodulation is coherent, a coherent carrier reference must be generated prior to demodulation. The generation of coherent reference at low signal-to-noise ratio is difficult. This difficulty is from the fact that any communication system must convey information from transmitter to the receiver. This implies that the carrier is in some way modulated with this information. In contrast to coherent case, the noncoherent detection is less complex and more robust against carrier phase variations. Sehier et al. [7] proposed to combine a linear equalizer with conventional differential detection, whereas Masoomzadeh-Fard et al. [8] considered decision-feedback equalizer with conventional differential detection. All of those schemes suffer from a significant loss in power efficiency compared to coherent case. Schober et al. [9] proposed the noncoherent minimum mean-square error (MMSE) receiver for DS-CDMA in a nondispersive channel. The receiver is related to the noncoherent linear MMSE equalizer reported in [10], i.e., the first stage of the receiver is a linear filter for interference suppression and the second stage is a decision-feedback differential detector. The receiver uses a certain observation window to generate a noncoherent decision variable. Varying the size of this observation window

1536-1276/08$25.00 c 2008 IEEE

k = 1. the filter taps are used to estimate a set of chip signals at one symbol time. so the length of the adaptive filter is the same as code length.1228 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS. and P Nk (t) is a wideband PN sequence defined by Lc −1 P Nk (t) = l=0 pk (l)Ψ(t − lTc ) where pk (l) ∈ {±1} is the l-th element of the PN code of user k. the base station transmits K data vectors of active users. Simultaneously. Finally. the number of samples in one symbol duration is DLc . The bandwidth of PN code is approximately 1/Tc . we apply the decision-feedback differential detection method in our proposed noncoherent receiver that is described in Fig. Θ denotes a constant phase shift introduced by channel. VOL. 1. where D is the number of samples during one chip duration in the receiver. The actual data rate may be varied due to the service provided or the actual transmission conditions. The CDMA downlink signal is then transmitted through a multipath channel. In our proposed receiver. In general. chip-level differential encoding/detection techniques for DS-SS signal to cope with frequency-nonselective fast fading channels were proposed. II. Therefore. APRIL 2008 U(n) Adaptive filter S C(n) q ( n) Reference symbol ^ d dif (n) d dif (n) ()* qref * (n . This paper is organized as follows.e. 0 ISI arises for Ln greater than one. the incoming waveform is downconverted in quadrature and sampled at the Nyquist rate.. conclusions are drawn in section VII. In [9].. K. S YSTEM M ODEL In the CDMA system. M − 1)}.i. The constant phase shift Θ is assumed to be known perfectly. III. we estimate the code delay τ0 to achieve code synchronization. 2. (3) where Lw is the number of taps of the transversal filter used in the receiver. which consists of independent and identically distributed (i.1) + PN sequence generator + Each data symbol bk (n) of user k is first oversampled by a spreading factor Lc and the resulting vector is multiplied element-by-element by the elements of PN spreading code sequence {pk (l)}. 4. Section V analyzes the performances of these proposed schemes. In addition to above process. the taps of the adaptive filter act as a despreader for code adjustment without considering a multipath channel. The received sample sequence {ri } can be expressed as [7][9][10] Ln −1 Ln −1 ri = e jΘ l=0 hl Si−l +υi = e jΘ l=0 hl S ((i − l)Td − τ0 )+υi (2) where S(t − τ0 ) means that S(t) suffers from timing offset.. In Section IV.d. NO. In [11] and [12]. The PN code samples generated in the receiver are represented by {ci } which is obtained by sampling P Nk (t) and ci = P Nk (iTd −τ ) where τ is the code delay which can be adapted to track the true code delay τ0 . the discrete-time channel is modeled relative to the bandwidth of the DS waveform. k = 1. 7. Some simulation examples are presented to demonstrate the effectiveness of the proposed receivers in section VI. the proposed noncoherent receiver with PN code tracking is presented. bk (n) ∈ {ej2πν/M | ν ∈ (0. [9] is the symbol-level signal estimation algorithm. Because the computational loading for chip-level differential detection is heavy in both encoding and decoding..i.d. A multipath channel can be represented by a tapped-delay-line with spacing equal to 1/DB.. The proposed noncoherent differential detection receiver with PN code tracking. we select the symbollevel differential encoding as in [9]. 1. (4) .. in [13]. In other words. . the decision feedback differential detection for DS-CDMA with multi-chip differential encoding was proposed to enable a large performance gain... Each user. the channel can be modeled as an FIR filter of length Ln whose impulse response is {hi }Ln −1 . we use the LMS algorithm to estimate the code delay through the same error signal. T HE C ONVENTIONAL C OHERENT R ECEIVER The conventional coherent receiver with PN code tracking is shown in Fig. In the same time. i.. This noncoherent detection requires the differential encoding in the transmitter and the reference symbol is needed for decoding in the receiver. The tap weight vector is W (n) = [w0 (n) w1 (n) · · · wLw −1 (n)]H .. DLc since the We assume that the channel order Ln maximum delay spread of the channel is usually insignificant in relative to the symbol period Ts [14]. k.. where B is the signal bandwidth. . In the digital DSCDMA receiver. which consists of Lc in general complex chips at chip interval Tc and satisfies the relation Tc = Ts /Lc . K. and τ0 denotes the true code delay. The noise term {υi } is an 2 i.. Td = Tc /D. The Nyquist sampling interval is Td . Our problem is to design an adaptive filter {wi } that estimates (or predicts) the desired signal. The conventional coherent receiver with PN code tracking is given in section III. For our structure. The transmitted baseband spreadspectrum signal is defined as K−1 ∞ code delay estimator S(t) = k=0 n=−∞ bk (n)P Nk (t − nTs ) (1) Fig. . transmits a data symbol sequence {bk (n)}. while our proposed scheme is the chip-level signal estimation algorithm.) M -ary PSK symbols at symbol interval Ts . and Ψ(t) is the chip waveform. the adaptive filter is used to estimate the desired signal and suppress the multipath and MAI effects based on the received signals. we first describe the DS-CDMA multiuser communication system model. and MAI arises due to channel distortion or non-orthogonal spreading code or both. can provide gain in the power efficiency. Also. the noncoherent receiver employs the decisionfeedback differential detection to recover the MDPSK signal.. 2.. For this reason. Gaussian random sequence with the variance σn . The data symbol sequences of different users are independent. In section II. where bk (n) is the information-bearing symbol of the k-th user. The local PN code vector is C(n) = [cDLc (n−1)+1 cDLc (n−1)+2 · · · cDLc n ]T . Thus.

Note that for N = 2. H where μ is the step-size. The resulting MDPSK symbols bk (n) are given by bk (n) = ak (n)bk (n − 1) (13) Ψ2 (δ + lTd ) . the maximum correlation of PN sequence is DLc . (4). where 0 ≤ δ ≤ Td if Ψ(t) is time-limited. respectively. After filtering. At the receiver. and the received sample matrix U (n) can be described as equations ˆ result of dcoh (n). and can be represented as error signal. is the number of despreader output symbols ∂τ used to calculate ddif (n). Let ecoh (n) = dcoh (n) − dcoh (n) be the (3). To achieve this goal we can use the cost function Jcoh = 2 ˆ E dcoh (n) − dcoh (n) where dcoh (n) is the hard decision . the cost function Jcoh can be written as q(n) = W H (n)U (n)C(n)/β. The i. The value β depends on the sampling of the chip waveform Ψ(t).UENG et al. a significant performance improvement can be obtained. the sampled signals are first passed through the transversal filter and then despreaded by the local PN sequence. and (5).. . for N > 2. cDLc n def ˆ ∂τ ∂τ ∂τ ∂τ defined as edif (n) = ddif (n) − ddif (n). The quan∂τ tity − ∂Jcoh can be regarded as an ”error signal”. ··· ··· rDLc n rDLc n−1 . Here. estimating ∂τ the chip-timing error of a code tracking loop. ddif (n) = ∗ q(n)qref (n − 1) where the reference symbol qref (n − 1) is where (·)∗ denotes complex conjugation. We can use the cost function where 2 ˆ ddif (n) − ddif (n) . In the next stage. we need 2 to compute ∂Jcoh = − β e∗ (n)U (n)C(n) and coh ∂W N −1 l−1 1 T q(n − l) ddif (n − m) (14) qref (n − 1) = ∂Jcoh ∂C (n) H N −1 = −dcoh (n) U (n)W (n)/β m=1 l=1 ∂τ ∂τ ∂C(n) /β −d∗ (n)W H (n)U (n) coh where N. β is no longer an integer. Define the sample matrix ⎡ ⎢ ⎢ ⎢ ⎣ U (n) = rDLc (n−1)+1 rDLc (n−1) . (5) ⎤ ⎥ ⎥ ⎥ ⎦ The tap weight of the adaptive filter is thus updated by W (n + 1) ∂Jcoh ∂W 2 ∗ e (n)U (n)C(n) = W (n) + μ β coh = W (n) − μ (12) ··· rDLc (n−1)−(Lw −2) rDLc (n−1)−(Lw −3) · · · rDLc n−(Lw −1) def ˆ Y (n) = [yDLc (n−1)+1 yDLc (n−1)+2 · · · yDLc n ]T ˆ The estimated signal Y (n) is denoted by (6) = U (n)W (n). The decision variable ddif (n) is obtained by noncoh +W H (n)U (n)C(n)C T (n)U H (n)W (n)/β 2 ] (8) coherent processing of the despreader output q(n). That is. the differenJcoh = E[dcoh (n)d∗ (n) − dcoh (n)C T (n)U H (n)W (n)/β tial detection is necessary to recover the MDPSK information coh −d∗ (n)W H (n)U (n)C(n)/β sequence. However. ν ≥ 1. = = pk (l) Ψ(iTd − τ − lTc ) ∂τ ∂τ ∂τ these past tap weight vectors are treated as constants since l=0 2 (11) |edif (n)| is differentiated only with respect to W (n). The error signal can be T Jdif = E ∂ ∂ ∂ ∂C(n) = cDLc (n−1)+1 cDLc (n−1)+2 . ddif (n) is the hard decision result ∂C(n) T ∂C T (n) H + W (n)U (n) C (n) + C(n) of ddif (n).d.: ADAPTIVE DS-CDMA RECEIVER WITH CODE TRACKING IN PHASE UNKNOWN ENVIRONMENTS 1229 where [·]T denotes Transposition and [·]H denotes Hermitian operation. β = Lc × max δ D−1 l=0 IV. with sampling time Tc /D and without filtering. information sequence {ak (n)} is first differentially encoded. The received sample sequence {ri } is also expressed as (2). ∂τ ∂τ ddif (n) is the decision variable of a conventional differential × U H (n)W (n)/β 2 (9) detection. The code delay τ is updated through τ (n + 1) = τ (n) − λ ∂Jcoh where λ is the step-size. Here. 2 shows the block diagram of the proposed noncoherent receiver that combines the differential detection with PN code tracking for DS-CDMA systems. that is.i. T HE P ROPOSED N ONCOHERENT R ECEIVER Fig. For the derivation of the adaptive algorithm. rDLc (n−1)+2 rDLc (n−1)+1 . It is desired that the output of the adaptive filter is the estimation of the desired signal. When the LMS generated [10] as follows algorithm is used to minimize the cost function Jcoh . the constant phase shift Θ is unknown. . so that the normalized despreader output is ˆ dcoh (n) = W H (n)U (n)C(n)/β (7) and the transmitted signal model is the same as (1). . local PN code vector C(n).. . edif (n) at (10) the n-th symbol time also depends on past tap weight vectors Lc −1 ∂ci ∂P Nk (iTd − τ ) ∂ W (n−ν). In other words. Note that C(n) has the property C(n)T C(n) = β where β is a constant that can be determined. . but still selected by maximum correlation value in general. . The tap weight vector W (n). we want that the signal at the despreader output to have the same statistic as that of an ideal MPSK demodulator output. The normalized despreader output q(n) is the same as (7). The and the output behaves like an MPSK signal. N ≥ 2. qref (n − 1) = q(n − 1).

. in the presence of multipath. pk (1) ⎢ ⎢ . In the nondispersive situation. the n-th received vector could be regarded as r (n) = rLc (n−1)+1+id rLc (n−1)+2+id · · · rLc n+id T (22) For the adaptive algorithm.. We assume that dcoh (n) = b0 (n). The tap weight ∂τ of the noncoherent adaptive filter is updated by 2 ∗ q (n − 1)e∗ (n)U (n)C(n) W (n + 1) = W (n) + μ dif β ref (18) ∂Jdif and the code delay τ is updated by τ (n + 1) = τ (n) − λ ∂τ . However. the optimum solutions for coherent and noncoherent detection are both described. the effective signature waveforms of active users are . ⎢ . Lc + Ln − 2. For simplicity. but only the MAI and interchip interference (ICI) to be considered. .. the baseband received signal for downlink CDMA communication can be represented as [15] K−1 ∞ where N (n) is the noise vector. 7. NO. . The received signal introduced by the n-th transmitted symbol is K−1 r(n) = ejΘ k=0 αk bk (n) + N (n) K−1 (21) = ejΘ {α0 b0 (n) + k=1 αk bk (n)} + N (n) The gradient vector ∂C(n) is the same as (10). ⎢ pk (Lc − 1) . that is D = 1. . ⎢ . A. Also. i = 0. and thus are unknown to the receiver. VOL. pk (Lc − 1) ⎥ ⎥⎡ ⎤ ⎥ h0 ⎥ ⎥⎢ . under steady-state condition. ⎥ ⎥⎢ . V. Under the situation D = 1. Optimum Solution of Coherent Detection In the first step. we assume that the decision variable dcoh (n) and ddif (n) are both perfect and C(n) = C in steady-state. The discrete-time counterpart of Ln −1 l=0 the signature waveform is given by αk (i) = hl pk (i − l). it can be shown that the performance of the proposed noncoherent CDMA receiver approaches that of the coherent one.. pk (Lc − 1) . Note that r(n) is an (Ln + Lc − 1) × 1 vector. APRIL 2008 cost function of differential detection Jdif can be written as Jdif = E[ddif (n)d∗ (n) (15) dif qref (n − 1) T C (n)U H (n)W (n) −ddif (n) β ∗ qref (n − 1) H −d∗ (n) W (n)U (n)C(n) dif β 2 |qref (n − 1)| + W H (n)U (n)C(n)C T (n)U H (n)W (n)] β2 The gradient of the cost function with respect to the tap weight vector is ∂Jdif 2 ∗ = − qref (n − 1)d∗ (n)U (n)C(n) dif ∂W β 2 2 |qref (n − 1)| U (n)C(n)C T (n)U H (n)W (n) + β2 2 ∗ = − qref (n − 1)e∗ (n)U (n)C(n) (16) dif β and the gradient of the cost function with respect to the code delay is ∂Jdif ∂τ qref (n − 1) ∂C T (n) H = −ddif (n) U (n)W (n) β ∂τ ∗ qref (n − 1) H ∂C(n) −d∗ (n) W (n)U (n) dif β ∂τ 2 |qref (n − 1)| + W H (n)U (n) β2 × ∂C(n) T ∂C T (n) C (n) + C(n) ∂τ ∂τ U (n)W (n) (17) H functions of propagation channels. ⎢ ⎢ . ⎢ ⎢ . ⎥ ⎥ hL −1 n ⎥ ⎦ For simplicity. ⎣ . let b0 (n) be the desired signal. . . αk (t) = Ln −1 l=0 hl pk (t − lTc ). The effective signature waveform αk (t) is related to the user’s spreading code and channel impulse response. 4.1230 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS. Because of the channel delay. . . then ∂Jcoh = E [e∗ (n)U (n)C] = 0 coh ∂W ⇒ E [U (n)b∗ (n)] C = 0 1 E U (n)CC T U H (n) W β (23) (24) r(t) = ejΘ k=0 n=−∞ bk (n)αk (t − nTs ) + υ(t) (19) If we define the following parameters Φcoh = E U (n)CC T U H (n) ϕcoh = E [U (n)b∗ (n)] C 0 Then the optimal weight W coh is derived as follows W coh = βΦ−1 ϕcoh coh (27) (25) (26) where αk (t) is the effective signature waveform of the kth user. . ⎤ (20) pk (0) pk (1) . the sample interval of pk (·) is Tc .. ⎢ . assume that there is only one sample per chip duration. ⎥ ⎥⎢ . we must derive the optimal weight W coh such that the MSE is minimized. The matrix formulation of αk (i) can be represented as ⎡ αk = pk (0) ⎢ pk (1) pk (0) ⎢ . Because of the Lc . Thus. ⎥ ⎥⎢ . For an infinite number of feedback symbols.. E[W (n)] = E[W (n − 1)] = W holds. P ERFORMANCE A NALYSIS In the following. the receiver signal can be seen as assumption of Ln an ISI-free signal. ⎥ ⎥⎣ . the user’s spreading waveform is the convolution of its assigned spreading codeword and the pulse shaping filter. ⎦ .

ddif (n) = a0 (n) can be assumed. and assuming that trivial solution ϕH W = 0 is excluded. (39). (26). H (38) gl + 1 β l=0 Lc −1 m=0 p0 (m)pk (m m=0 Lw −1 l=0 We multiply e−jΘ W e−jΘ W ϕdif = H with ϕdif and obtain Lc −1 p0 (m) wl υLc (n−1)+m+id −l Ln +Lw −2 gl l=0 m=0 p0 (m)p0 (m + id − l) = Δ. we find |Δ| = β 2 ϕH Φ−1 ϕdif . If perfect decision is achieved. then (33) can be reduced to 1 2 |Δ| E U (n)CC T U H (n) W = e−jΘ Δ∗ E [U (n)b∗ (n)] C 0 β2 (34) Here. Optimum Solution of Noncoherent Detection for Infinite Number of Feedback Symbols In order to explain the effect qref (n − 1) has in the limit as N → ∞ . 1)-th element of ϕdif has the form Ln −1 Lc −1 gl m=0 Lc −1 p0 (m)p0 (m + id − l) b0 (n) + id − l) bk (n)} (30) l=0 Ln +Lw −2 ejΘ l=0 hl m=0 p0 (m)p0 (id + m − (xw − 1) − l). E [U (n)b∗ (n)] is an Lw × 0 Lc matrix. the despreader output q(n) can be rewritten as 1 1 ˆH q(n) = Y (n)C = β β = + k=1 Lc −1 m=0 ∗ p0 (m)yLc (n−1)+m = ejΘ l=0 1 jΘ e { β K−1 Ln +Lw −2 Lc −1 ϕdif is an Lw × 1 vector and the (xw . β where Δ can be seen as a constant term that has no information about symbol data. (36). When (13) is applied (Note that a0 (n) is the information symbol). The estimated x-th chip at n-th symbol time as in (6) can be represented as ∗ yLc (n−1)+x Lw −1 ∗ E a∗ (n)qref (n − 1)U (n)C (33) 0 1 2 E |qref (n − 1)| U (n)CC T U H (n) W β 2 using (13). the first term is the desired signal term. the coherent MMSE J coh can be described as J coh = = 1− 1 H 1 H 1 H ϕcoh W coh − W coh ϕcoh + 2 W coh Φcoh W coh β β β 1 − ϕH Φ−1 ϕcoh (28) coh coh MAI term and noise term of reference symbol qref (n−1) can be ignored.: ADAPTIVE DS-CDMA RECEIVER WITH CODE TRACKING IN PHASE UNKNOWN ENVIRONMENTS 1231 Applying (25). xc )-th element can be written as E rLc (n−1)+id +xc −(xw −1) b∗ (n) 0 Ln −1 + l=0 wl υLc (n−1)+x+id −l (29) = ejΘ E l=0 K−1 hl pk (id + xc − (xw − 1) − l)bk (n) b∗ (n) 0 hl p0 (id + xc − (xw − 1) − l) (37) where gl = Lw −1 m=0 wm hl−m are the coefficients of the discrete- × k=0 Ln −1 time overall impulse response. we also define that Φdif = E U (n)CC T U H (n) ϕdif = E [U (n)b∗ (n)] C 0 (35) (36) = l=0 wl rLc (n−1)+x+id −l Ln +Lw −2 = ejΘ { K−1 l=0 Ln +Lw −2 gl p0 (x + id − l) b0 (n) gl pk (x + id − l) bk (n)} + k=1 Lw −1 l=0 Observing the matrix ϕdif in (36). and the (xw . dif dif 2 (42) . The noncoherent optimal weight can be derived from (16) ∂Jdif ∗ = E qref (n − 1)e∗ (n)U (n)C = 0 dif ∂W ⇒ = (32) Note that ΦH = Φcoh . coh B. the reference symbol (14) in the limit N → ∞ can be reduced to qref (n − 1) = = 1 jΘ e β Ln +Lw −2 Lc −1 (39) Using (35). (31). Therefore. dif β gl l=0 m=0 p0 (m)p0 (m + id − l) b0 (n − 1) (31) (40) (41) 1 jΘ e Δb0 (n − 1). the performance of noncoherent detection can be improved as N increased. the second term is the MAI term.UENG et al. Observing (30). (34) can be rewritten as dif 1 2 |Δ| Φdif W = e−jΘ Δ∗ ϕdif β2 ⇒ H 1 Φ W W ϕdif = ϕdif 2 dif β H 1 ⇒ Φ−1 = 2 W W . Due to the effect of ensemble average. the From (39) and (41). and the last term is the noise term. and (27) into (8). and the fact |b0 (n)| = 1. we must describe the received signal model in detail. For this reason.

(42). Except for Fig. (35). The 31 Gold code is generated by a set of polynomials z 5 + z 3 + 1 and z 5 + z 4 + z 3 + z + 1. Δ can be represented as Δ=β ϕH Φ−1 ϕdif ejφ0 . 3-10. This implies that the chip waveform is restricted to the duration Tc so that the neighboring chip will not interfere with each other. we use simulation to compare the effectiveness of different chip waveforms. the MMSE solution of noncoherent differential detection is the same as that of the coherent detection. (44) 10 −7 ACT with 30 users ACT with 20 users ACT with 10 users TWCT with 30 users TWCT with 20 users TWCT with 10 users 10 −8 1 2 3 4 5 6 7 Eb/No (dB) Note that W dif is unique only up to a complex factor with magnitude one. we compare the adaptive code tracking loop (ACT) used in our receivers with code tracking scheme by observing tap weight (TWCT) in [2]. Thus. (36). In this section. 3. The number of samples per chip D is 4 or 5. it is generally difficult to evaluate this effect. and (43). and channel 2 can be described as h2 = [ 1 0 0.005 and λ as 0. and (39) into (15). we show the mean square tracking error. The adaptive filter length is 15 taps. E |τ |2 . The phase shift Θ is uniformly distributed. dif dif (43) Mean Square Tracking Error Regular Timing Delay in Channel 2 10 0 10 −1 where φ0 is an arbitrary phase. 1 2 3 and ε3 = 0. However. waveform is commonly used. we have W dif Φdif W dif = β 2 . the difference between optimal weights W coh and W dif are the arbitrary phase and the scaled magnitude term. (40).08 2 [1 − cos(2πt/Tc )] ζc (t) 3 equal to Tc . we normalize the desired signal power of the channel output by setting the channel impulse response as h/ h . 3) Half-Sine (hs) : Ψ(t) = √ 2 sin(πt/Tc )ζc (t) From (44). the optimal weight of noncoherent detection W dif can be calculated as ⇒ e−jΘ W ϕdif = β H H 10 −2 10 −3 ϕH Φ−1 ϕdif ejφ0 dif dif ϕH Φ−1 ϕdif ej(Θ+φ0 ) dif dif 10 −4 10 −5 ⇒ Φdif W W ϕdif = βΦdif W ⇒ βϕdif = Φdif W ⇒ W dif = βΦ−1 ϕdif dif ϕH Φ−1 ϕdif dif dif 10 −6 ϕH Φ−1 ϕdif ej(Θ+φ0 ) dif dif e−j(Θ+φ0 ) . 9. the performance of noncoherent differential receiver for MDPSK approaches the conventional coherent receiver for MPSK. dif dif β2 β (45) Fig. and (46). H where ζc (t) = 1 for 0 ≤ t ≤ Tc and ζc (t) = 0 otherwise. where τ = τ −τ0 . We use a Gold sequence for the spreading code on the forward channel. 5. Comparing (27) and (44). QPSK for coherent detection and DQPSK for noncoherent detection are discussed because of the higher bandwidth efficiency. then (45) can be reduced to 1 2 Jdif = 1 − 2 |Δ| = 1 − ϕH Φ−1 ϕdif .10-11. 3 and 4. Two channels are selected for our simulations.2194 0 1 0 0. (47) dif dif β Comparing (28) and (47). In Figs.42. The adapted code delay τ is initialized with 0.005. the following time-limited chip waveforms are considered :[16][17][18] 1) Raised-Cosine (rct) : Ψ(t) = 2) Blackman (bm) : Ψ(t) = [ε1 − ε2 cos(2πt/Tc ) + ε3 cos(4πt/Tc )] ζc (t) where 2 = (ε2 + ε2 /2 + ε2 /2)−1 and ε1 = 0. In general. From (39). we choose L = D in most case. with D = 5 in channel 2 for both code tracking schemes. These chip waveforms are normalized to have energy Tc (46) Using (39). Note that the tap weight length of TWCT is 15 which is the same as that of our proposed receiver. All learning curves are simulated in Eb /N0 = 10 dB as in Figs. |e(n)| versus n. VI. L is related to the path delay. 4. the noncoherent MMSE solution Jdif can be written as Jdif = 1 − H H H 2 1 W ϕdif ϕH W + 4 W ϕdif W Φdif W ϕH W . This will be shown by computer simulations later. Applying (31). We choose the step-size μ as 0. for the infinite number of feedback symbols. APRIL 2008 Thus. S IMULATION R ESULTS From the above discussion. 7.5 0 0. the raised-cosine . we know that the selection of the chip waveform and the sample number D are the factors to control the performance of the system.5. First. that is the received timing of each path is separated by one chip duration. over 1000 independent trials. The impulse response of the channel 1 is described by raisedcosine h1 = [ 0.2194 ]T . Instead of binary modulation. In the simulations. The learning curve of error signal is obtained by averaging the absolute value of instantaneous-error. so we focus our simulations on it in Figs. ε2 = 0. The effect of the gradient of chip waveform about the code delay as in (11) on the performance of the system can be large. Therefore.1232 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS. where 0 is an 1 × L zero vector. VOL. The learning curve of timing error is the 1000 independent trials average for τ .2 ]T . Comparison of mean square tracking error for ACT and TWCT code tracking algorithm for different number of users. 0 Ψ2 (t)dt = Tc . NO. as the dependence of the system performance on the chip waveform is usually very complicated.

we . This might be attributed to the fact that the reference symbol qref (n − 1) always has the ICI term even if N → ∞.75 Tc in Fig. TWCT has better tracking effect than ACT. In Fig. our proposed receiver can effectively track the PN code timing offset. So we use an irregular timing delay with τ0 = − 1.: ADAPTIVE DS-CDMA RECEIVER WITH CODE TRACKING IN PHASE UNKNOWN ENVIRONMENTS One User in Channel 2 10 −1 1233 Channel 1 10 −2 −1 10 Mean Square Tracking Error 10 −2 10 −3 SER 10 −4 10 −3 10 −5 ACT with irregular timing delay ACT with regular timing delay TWCT with irregular timing delay TWCT with regular timing delay 10 −4 noncoherent N=2 with hd noncoherent N=3 with hd noncoherent N=5 with hd noncoherent N=10 with hd noncoherent N=10 with ps coherent with hd coherent with ps 2 3 4 5 6 7 8 9 10 10 −6 Eb/No (dB) 1 2 3 4 5 6 7 Eb/No (dB) Fig. 2 In Fig. When the number of active users approaches the processing gain of spreading code. where K = 20 and τ0 = Tc /2. But this regular timing delay situation occurs with very low probability. 6. the SERs of the conventional coherent receiver are also shown. Furthermore. Therefore. where K = 10 and τ0 = Tc /2. This property causes the inherent tracking error when the timing delay is irregular. Eb /N0 for the proposed noncoherent receivers. In Fig.s. 8. and the performance of noncoherent receiver can approach that of the coherent receiver further. For comparison. Because the TWCT works 5 by observing the peak of tap-weight vector. the TWCT has poor performance than ACT (especially for K = 30). Figs. the learning curves for these receivers are given. In the above simulations. 5. the MAI term of reference symbol qref (n − 1) will be strong and the larger number of feedback symbols is necessary to facilitate the effect of ensemble average. When the number of users is small. 6-9. Learning curves of |e(n)| for noncoherent receiver with different N and coherent receiver. Second. Case 4 : The conventional coherent receiver with perfect synchronization (ps). we compare the SER performances of our proposed noncoherent receiver with the conventional coherent receiver under the following cases: Case 1 : The proposed noncoherent receiver in half chip duration timing offset (hd). we compare the performance of noncoherent receiver with that of the conventional coherent receiver with D = 4. Channel 2 10 0 10 10 −0. a regular timing delay with τ0 = − 5 Tc is used.1 −1 10 −0. where Eb /N0 = 10 dB and τ0 = Tc /2.3 N=3 N=5 −0. so it can only track timing offset with a multiple of sampling time. Case 3 : The proposed noncoherent receiver (N = 10) with perfect synchronization. SER v. symbol-error-rates (SERs) in steady state are shown. the SERs of the conventional coherent receiver are also shown. Eb /N0 for the proposed noncoherent receivers. Comparison of mean square tracking error for ACT and TWCT code tracking algorithm for regular and irregular timing delay. Clearly. the performance of the proposed noncoherent receiver can approach the performance of the conventional coherent receiver as N grown. 4.s. The noncoherent receiver spends longer time to deal with these terms. In Figs. Case 2 : The conventional coherent receiver in half chip duration timing offset (hd).2 N=2 |e(n)| 10 −0.6 Coherent 0 100 200 300 400 500 600 10 −3 noncoherent N=2 with hd noncoherent N=3 with hd noncoherent N=5 with hd noncoherent N=10 with hd noncoherent N=10 with ps coherent with hd coherent with ps 2 3 4 5 6 7 8 9 10 number of iterations (n) Eb/No (dB) Fig. 7. SER v. Note that the convergence speed of the coherent receiver is faster than that of the noncoherent receiver.5 10 10 −0. In these two figures. 6 and 7 show the SER versus SNR with 10 and 20 active users. TWCT has poor performance even if the number of active user is 1.4 SER 10 −2 10 N=10 −0. and have almost the same performance as that of the receivers with perfect PN code synchronization.UENG et al. Therefore. For comparison. qref (n−1) will approximate the form of (31). However. when the number of users is large. respectively. 4. Fig. the SERs of different number of active users are compared at Eb /N0 = 10 dB. Channel 2 Fig. 3. 5.

s. 9. C ONCLUSION A novel noncoherent receiver for joint timing recovery and data detection in DS-CDMA systems is proposed in this work. The loss of noncoherent detection compared with conventional coherent detection is limited and can be adjusted via the generation of the reference symbol for the decision-feedback differential detection. Note that the tracking performance of the raised-cosine and Blackman waveforms are better than that of the half-sine waveform in both tracking speed and stability.4 hs −0. Three different chip waveforms are simulated in two different multipath channels with different numbers of active users. 10. VOL. where Eb /N0 = 10 dB and τ0 = Tc /2.6 0.8 10 −3 0 500 1000 2 3 4 5 6 Eb/No (dB) 7 8 9 10 1500 number of iterations (n) 2000 2500 3000 Fig.2 t racki er ng ror t 0 = Tc / 8 0 SER 10 −3 -0. As shown in Figs. The phase shift of each path is generated by uniform distribution independently.s. each pulse is no longer separated by one chip duration. Fig. where K = 25 and τ0 = Tc /2. In fact.2 t 0 = -Tc / 4 t 0 = -Tc / 8 10 −4 noncoherent N=2 noncoherent N=3 noncoherent N=5 noncoherent N=10 coherent -0. 4. after tracking process. 9. the phase shift of each path is generated independently. 6 and 7. the phase shift introduced by each channel path is the same. 0 Fig.4 10 −2 t 0 = Tc / 2 t 0 = Tc / 4 0. In Fig. the tracking error can converge to almost the same value in each case. It is shown that the timing offset can be rapidly tracked even if the mismatch is up to half chip time interval. 11. VII. Learning curves of τ different τ0 . It estimates the desired signal and code delay by LMS algorithm at the same time. 7 and 9. K for the proposed receivers. The performance of the noncoherent receiver with N = 10 is very close to that of the coherent receiver. so we can anticipate that the SER performances of these three different waveforms are nearly equal. In Fig. the performances of three different chip waveforms are compared. The MMSE solution of the proposed receiver is analyzed theoretically and by computer simulations. SER v. simulations show that the proposed receiver in an asynchronous situation .2 10 −2 −0. the noncoherent receiver is more popular than coherent one.4 0. Eb /N0 for the proposed noncoherent receivers.1234 Channel 1 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS.4 t 0 = -Tc / 2 -0. Observing Fig. the trend of the performance comparison in Fig. 8.6 for N = 10 noncoherent receiver with Channel 2 channel 2 (each path suffers from different phase shift) noncoherent N=2 with hd noncoherent N=3 with hd noncoherent N=5 with hd noncoherent N=10 with hd noncoherent N=10 with ps coherent with hd coherent with ps 10 0.8 0 500 1000 number of iterations (n) 1500 Fig. that is h = [ h0 ejΘ h1 ejΘ · · · hLn −1 ejΘ ]T . Every situation can be rapidly tracked. the asynchronous situation can achieve almost the same SER as the perfect synchronization even if a little tracking error still exist. The performance of the noncoherent receiver can approach the performance of the conventional coherent receiver if an infinite number of feedback symbols is used. 0. SER v. for N = 10 noncoherent receiver with set the channel model under the assumption of (2).2 rct hs 10 −1 tracking error 0 bm rct bm SER −0. 10 with D = 4. 7. 11. Different timing offset is shown in Fig. Observing (2). In this channel model. 6. Furthermore. Besides. 9 is the same as that in Figs. 11. and facilitate the signal error |e(n)| to converge.6 −0. Learning curves of τ different chip waveforms. we choose the noncoherent receiver with N = 10 to explain the tracking effect.6 10 −5 5 10 15 number of users (K) 20 25 30 -0. the simulation channel is channel 2 and L = D + 2. NO. Therefore. as has been shown analytically. that is. APRIL 2008 Channel 1 10 −1 0.

“Chip-level differential encoding/detection of spread-spectrum signals for CDMA radio transmission over fading channels. Taichung. 2325– 2329. [18] R.-J. El-Tarhuni and A. smart antenna.E.. fading. Oct. 413–424. 2001.. 1036–1045. and communication signal processing. “DF-DD for multi-chip differentially encoded DS-CDMA.E and M. Commun. Shang-Chun Tsai received the B. [5] R. 44. [8] A. Liu and K. Areas Commun. H.E degree form Da-Yeh University. G. R. 47. pp.” IEEE Trans. U. and H. “Performance analysis for an adaptive filter code-tracking technique in direct-sequence spreadspectrum systems. Wireless Commun.S.” IEEE Trans. J. 512–520. [4] P. pp. vol. G. 672–675. Wong. “Low complexity adaptive code tracking with improved multipath resolution for DSCDMA communications over fading channels. Liu and G.1677–1685. Nguyen and E. Gerstacker. vol. “A robust PN code tracking algorithm for frequency selective Rayleigh-fading channels.. Hsinchu. .T. degree in electronic engineering from the National Chiao Tung University. Signal Processing. “On error probabilities of DS-CDMA systems with arbitrary chip waveforms. From 1996 to 2001 he was with National Space Program Office (NSPO) of Taiwan as an associate researcher.. His research interests include multiuser detection. “Closed-loop least mean square time-delay estimator. Schober. pp. Pasupathy. pp. “Adaptive linear equalization combined with noncoherent detection for MDPSK signals. no.” in Proc. Jan. Lampe. “Joint estimation of PN code delay and multipath using the extended Kalman filter. 35. 1990. H. vol. 10. Wyrwas.” in Proc. “The effect of chip waveform on the performance of CDMA systems in multipath. B. [3] M. Reggiannini. 50. 38. Gerstacker. 4. Commun.UENG et al. no. Commun. Raheli. pp. Apr. Fang-Biau Ueng received the Ph. R.” IEEE Commun. pp. Lehnert. pp.D. Schulz-Rittich.C. vol. Taiwan in 1995. 2003. 30–34. H. vol. pp. 103–114. H.: ADAPTIVE DS-CDMA RECEIVER WITH CODE TRACKING IN PHASE UNKNOWN ENVIRONMENTS 1235 approaches the performance as that of the receivers with perfect synchronization.E. Commun.. Hosemann. Schober and L.. Taiwan. El-Tarhuni and A.. He is currently a postdoctoral researcher in the Department of Electrical Engineering at National Chung-Hsing University. Mar. 1999. W. Acoustics. 14. Speech. 2000.” IEEE Trans. pp. degree in electrical engineering from the National Chung-Hsing University. M. Sheikh. pp. Commun. 5. Jun-Da Chen received the Ph. K. Apr. 48.S. 1990. and J. 1996.” in Proc. Apr. Sept. Taiwan. “Noncoherent MMSE interference suppression for DS-CDMA. vol. Dec. 137. Li. Select. 1997. Commun.” IEEE Trans. in 2001. Masoomzadeh-Fard and S. Messer and Y. July 1999. Jan. National Chung-Hsing university. 46. His research interests include the areas of MIMO signal processing. vol. IEEE Sixth International Symposium on Spread Spectrum Techniques and Applications. degree in the Department of Electrical Engineering at National Chung-Hsing University. spread spectrum CDMA. vol. Aug.” IEEE Trans. noisy channels. 2000. 125–133. Huber. 733–738. “Nonlinear equalization of multipath fading channels with noncoherent demodulation.. Taiwan. IEEE V. May 2000. H.O. 4. Taichung. Baltersee.” IEEE Trans. 4. pp. From 2001 to 2002 he was with Siemens Telecommunication Systems Limited (STSL). and R. Feb. [14] H. Srikanteswara. smart antennas. T. “A subspace method for signature waveform estimation in synchronous CDMA systems.. [13] R.” IEE Proc. Shwedyk. pp.. Sept. 78–80. 1996. 1998. F. 577–587. His areas of research interest are wireless communication and adaptive signal processing. R EFERENCES [1] M. Xu. [9] R. Giannetti. 9–12. vol. pp. 456–463. 1018–1023.. Oct. no. pp. vol. Kaleh. [11] A. Commun. 1987. vol. 1. Wireless Commun. and 2003.. Lok. pp. [10] R. [2] M. vol. A. Cavallini. F. Since February 2002 he has been with the department of electrical engineering. 2002. Anjaria and R.” IEEE Trans. [7] P.” in Proc. IEEE Radio and Wireless Conference. “A code tracking technique for direct sequence spread spectrum using adaptive filtering.” IEEE Trans. Ghrayeb. Reed. Taipei. IEEE GLOBECOM.” IEEE J. Lampe. vol. S. [12] G. Fock.. vol..2.D. “Adaptive equaliser for differentially coherent receiver. May 1992. and J. [16] T. Commun. vol. Taiwan in 2006. He is currently working toward the Ph.” IEEE Trans. S. 25–28. vol. pp. Lett. Bar-Ness.” IEEE Trans. “Improved differential detection of chiplevel differentially encoded direct-sequence spread-spectrum signals. H. Taiwan. Luise. [15] H. 1. Apr. Conference. “Asynchronous multiple-access interference suppression and chip waveform selection with aperiodic random sequences. M. 45. [6] H. 47. pp. “A decorrelating RAKE receiver for CDMA communications over frequency-selective fading channels.. Schober. 1058–1064. 1346–1354.” IEEE Trans. pp. Sehier and G. and A. July 2004. 3. 2002. Iltis. and J. Colavolpe and R. W. vol.C.O. respectively. where he was involved in the design of mobile communication systems. wireless communication systems and multiuser communication systems. [17] H. Meyr.D. Commun.

Sign up to vote on this title
UsefulNot useful