SIGNAL DESIGN FOR LS AND MMSE CHANNEL ESTIMATORS Pertti Järvensivu, Marja Matinmikko, Aarne Mämmelä VTT Electronics, Kaitoväylä 1, P.O. Box 1100, FIN-90571 Oulu, Finland, Pertti.Jarvensivu@vtt.fi Abstract - We analyzed the effect of time and frequency domain windowing, or weighting, of the transmitted signal for least squares (LS) and minimum mean-square error (MMSE) estimators when the channel is time-variant. We considered the estimation error for different windows and we found that the windows can be selected independently. We explained the performance of the estimators with the help of the radar ambiguity function of the transmitted signal including the windowing. Keywords - Doppler-delay-spread function, matched filter bank, square-root raised cosine, Hamming, Blackman. I. INTRODUCTION In this paper, we analyze the effect of time and frequency domain windowing of the transmitted signal on the perfor- mance of least squares (LS) and minimum mean-square error (MMSE) estimators. We also present some simulation results. We use Bello’s [1] estimator of the Doppler-delay- spread function, which is the Fourier transform of the output delay-spread function with respect to the time variable [2]. Bello’s work was an extension of Levin’s [3] work in time- invariant channels to time-variant channels. Bello’s estimator includes a matched filter bank, a set of samplers and an equalizer for the sidelobes of the ambiguity function of the transmitted signal (Fig. 1). The estimator can be easily extended to estimation of the output delay-spread function by using a set of inverse discrete Fourier transforms [4]. The MMSE estimator corresponds to the estimator pres- ented in [5], which estimates the input delay-spread function (also known as the time-variant impulse response) instead of the output delay-spread function. A comprehensive review of channel and other estimators was presented in [6] where the relationships between different channel estimators are clearly shown. An earlier review is in [7] which includes Levin’s estimator. Implementation of a matched filter bank with the discrete Fourier transform (DFT) was considered already in [8]. DFT can be used if the measurement signal is effectively a rectangular pulse without any modulation. The use of fast Fourier transform (FFT) in filter banks is well known in the radar literature [9]. For example, Yao and Cafarella [10] used a simplified version of Bello’s [1] estimator and the DFT. They also used a Hamming window in the time domain in the receiver. However, no mathematical analysis was given to consider the effects of windowing. Some theory on time and frequency domain windowing, or weighting, of a linearly frequency modulated chirp pulse is presented in [11]. The organization of the rest of the paper is as follows. In Section II, we present the system model including the definition of the Doppler-delay-spread func- tion. We discuss factors affecting the system performance in Section III. In Section IV, we present some simulation results. Finally, we make some conclusions in Section V. Matched filter bank Equalizer r vˆ (a) Matched filter Matched filter Matched filter Equal- izer Matched filter bank Sampling r vˆ (b) Fig. 1. (a) LS and MMSE estimators of Doppler-delay- spread function (b) Matched filter bank. II. SYSTEM MODEL In this paper, we consider random time-variant channels. The complex envelope of the received signal can then be written as [2] (Sections III-B and D)       t n d d V e t z t n d t h t z t n d t g t z t r t j   , , , 2 (1) where z(t), 0 < t < T is the transmitted signal, g(t,) is the input delay-spread function, h(t,) is the output delay-spread function, V(,) is the Doppler-delay-spread function, is the Doppler shift, is the delay and n(t) is zero mean additive white Gaussian noise (AWGN). The transmitted signal z(t) is formed by first applying a single impulse or a sequence of impulses to a pulse shaping filter with an impulse response p(t), and then windowing the result with w(t) (Fig. 2). Although we consider m-sequences here, also other pseudorandom sequences could be used. 0-7803-7589-0/02/$17.00 ©2002 IEEE PIMRC 2002