MIMO Antenna Design and Channel ModelingView this Special Issue
Research Article | Open Access
Channel Parameter Estimation for Scatter Cluster Model Using Modified MUSIC Algorithm
Recently, the scatter cluster models which precisely evaluate the performance of the wireless communication system have been proposed in the literature. However, the conventional SAGE algorithm does not work for these scatter cluster-based models because it performs poorly when the transmit signals are highly correlated. In this paper, we estimate the time of arrival (TOA), the direction of arrival (DOA), and Doppler frequency for scatter cluster model by the modified multiple signal classification (MUSIC) algorithm. Using the space-time characteristics of the multiray channel, the proposed algorithm combines the temporal filtering techniques and the spatial smoothing techniques to isolate and estimate the incoming rays. The simulation results indicated that the proposed algorithm has lower complexity and is less time-consuming in the dense multipath environment than SAGE algorithm. Furthermore, the estimations’ performance increases with elements of receive array and samples length. Thus, the problem of the channel parameter estimation of the scatter cluster model can be effectively addressed with the proposed modified MUSIC algorithm.
It is important to estimate the spatial/temporal parameters, such as directions of arrival (DOAs), path delays, frequencies, and so forth, embedded in the receive signals in radar, sonar, and wireless communication systems. It also finds applications in source localization, accident reporting, cargo tracking, and intelligent transportation. For example, a precise estimation of DOAs and frequencies of rays in wireless communication channels can helpfully provide better channel information so as to enhance the performance in terms of coverage, capacity, and quality of service (QoS) considerably and increase resistance against interferences. On the other hand, an effective channel model must rely on a realistic characterization of the probability distribution of the relevant channel parameters. So the validation of channel parameter is a prerequisite to ensure that these models reproduce the critical features of the propagation environment, that is, in delay, direction, Doppler, and polarization.
Recently, various high-resolution methods have been proposed in mobile channel to estimate some of the parameters of impinging plane waves, that is, their complex amplitude, relative delay, incidence azimuth, incidence elevation, and Doppler’s frequency. These methods can be grouped into three of the categories : spectral estimation, parametric subspace-based estimation (PSBE), and maximum likelihood estimation (ML). The first mentioning category is the MUSIC (multiple signal classification) algorithm . The ESPRIT (estimation of signal parameter via rotational invariance techniques)  and unitary ESPRIT  methods are included in the PSBE techniques. Among the ML methods, the expectation-maximization (EM) algorithm and the SAGE (space-alternating generalized EM) algorithm  have perfect performance. Especially, SAGE algorithm has been applied for joint delay and azimuth estimation in time-invariant environments as well as for joint delay, azimuth, and Doppler’s frequency estimation in time-variant environments.
Unfortunately, the computational burden of the SAGE algorithm is high due to the necessity of nonlinear and multidimensional optimization procedure. Since Swindlehurst proposed several computational efficient algorithms for the estimation of the delays of a multiray channel and solved the spatial signatures (or DOAs) as a least square problem , many algorithms which can estimate the channel parameters by a 2D searching on the DOA-delay domain or DOA-frequency domain have been proposed such as JADE-MUSIC algorithm , TST-MUSIC algorithm , FSF-MUSIC algorithm .
With the development understanding of the wave propagation phenomena, the ray model is not suitable to interpret the parameter estimation results of a channel parameter estimator. This is due to the limited resolution of any wireless channel system. A number of radio channel models based on scatter clusters have been proposed in the literature. Many such models which are referred to as cluster delay line (CDL) models simplify the scattering environment and thereby precisely evaluate the performance of the communication system.
However, the SAGE algorithm did not work for the scatter cluster model because the transmit signals are highly correlated. In this paper, we present a low-complexity, yet high-accuracy, MUSIC-based algorithm, which combines the techniques of temporal filtering and of joint DOA and frequency with two-dimensional (2D) searching. Except for this, there are several other advantages of modified MUSIC algorithm compared with SAGE algorithm, such as the reduction of computation, reduced complexity.
This paper is organized as follows. Section 2 introduces the system model of the fading multiray channels, which assumes the propagation rays to be scatter cluster. In Section 3, we present modified MUSIC algorithm and some related issues. In Section 4, simulation results are presented to verify the performance of the proposed approach. Section 5 comes to the conclusions.
2. System Model
We consider a wireless communication system with closely spaced receive antennas , which is equipped with a uniform linear arrays (ULAs), and all the elements are omni-directional; the interelement spacing at the receive antennas is half a wavelength. Suppose that the signals have a common center frequency of , then the corresponding wavelength is , where is the speed of propagation. Generally, the radio channel in a wireless communication system is often characterized by a multiray propagation model. According to 3GPP SCM model , the received signal at the MS consists of time-delayed multipath replicas of the transmitted signal. These paths are defined by the same delays which are caused by scatter clusters, and each path consists of subpaths. Each subpath represents real ray. The total ray is . So, the output of the each element at the receive antennas () can be written as where denotes the th receive antenna response to the ray from , in which is the DOA of the th subpath of the th path. denotes Doppler’s frequency of the th subpath of the th path. If a narrow-band transmit signal and a slow-fading environment are considered, we can assume that the signals are block fading; that is, remains unchanged in a short period and fades independently from block to block. After appending time samples at the th receive antennas, we can obtain the block signals in matrix:
Using the general relation, , yields where , with denoting the sampling period and . Then the Khatri-Rao product , and is an additive noise process, assumed to be a zero-mean Gaussian noise vector with covariance . The receive array response matrix , , where denotes the array response vector of the th ray:
The frequency array response matrix is frequency samples in a short period, with denoting the sampling rate, : where denotes the transmit signal which has a delay .
3. Modified MUSIC Algorithm
Because the rank of correlation matrix is the number of cluster and not the number of ray, the conventional MUSIC algorithm does not work for scatter cluster model directly. So we make some modification based on the conventional MUSIC algorithm. The flow chart of the modified MUSIC algorithm is illustrated in Figure 1.
3.1. The Temporal Filter Method
Despite the conventional MUSIC algorithm constructs a spatial correlation matrix, the proposed algorithm is based on the decomposition of the theoretical temporal correlation matrix into a signal subspace and a noise subspace :
Note that the number of the incoming paths is assumed to be known a priori; otherwise, we may estimate by thresholding the magnitude of the eigenvalues of the covariance matrices involved or by using the AIC and the MDL detection methods , then the next step is similar to the conventional MUSIC algorithm:
The column vectors of are the eigenvectors that span the signal subspace , respectively, corresponding to the largest eigenvalues. The column vectors of spanned by the rest of the and - eigenvectors of are the orthogonal complement of the column vectors of .
Using the orthogonality property between the signal and the noise subspaces, the T-MUSIC algorithm estimates the path delays by
After searching over the range of interest, the spectrum of the T-MUSIC is, respectively, defined as
By applying the T-MUSIC, the resulting delays are estimated. Based on the delay estimates, we define the temporal filtering matrices as
Note that is the complement project matrix of with ; with these facts, the modified MUSIC algorithm postmultiplies to , which is referred to as the temporal filtering process to separate the rays with different delays. Then the output of the th temporal filter is given by
3.2. Spatial Smoothing (SS) Method
After applying the temporal filter, we can jointly estimate DOAs and Doppler frequencies of each delay cluster. However, the signals included a delay cluster are coherent, the matrix becomes singular, so that some of its eigenvalues are zero. This means that part of the signal subspace is indistinguishable from the noise subspace. As a result, the observed noise subspace is no longer orthogonal to the signal subspace and the MUSIC algorithm fails. So, to overcome these problems, we use a technique called spatial smoothing (SS) to allow the MUSIC algorithm to be applied to the coherent signal case .
The basic idea is to form covariance matrices from subsets of the array, which is equivalent to partitioning the original covariance matrix. If there are subarrays, each sub-array is of size and the output of forward subarray is denoted by with elements .
Then we can jointly estimate DOAs and frequencies in one delay cluster by the spatial correlation matrix , where
Similarly, after eigen decomposition of , the eigenvectors can be divided into two groups: eigenvectors corresponding to the largest eigen values are called signal eigenvectors and the subspace they span is called signal subspace. Utilizing the theory of orthogonality, we can obtain the spectrum:
4. Simulation Analysis
In this section, we conduct simulations to assess the proposed MUSIC algorithm. Assume narrow-band signals that are transmitted through 7 rays and received by a ten-element uniform linear array, the spacing of two elements is half a wavelength. The 7 rays are divided into two groups: 4 rays have the same delay of 20 μs, the remains are 40 μs delay; the first delay cluster: (80°, 30 kHz), (60°, 60 kHz), (40°, 90 kHz), (50°, 90 kHz) and the second delay cluster: (20°, 120 kHz), (45°, 30 kHz), (15°, 60 kHz). We sample the receive signal times, . In each short sample, we also have frequency samples, each signal’s SNR is set to be 20 dB, and the additive Gaussian white noise (AWGN) is considered.
Figure 2 indicates the two delay clusters estimation of the synthetic channel. Each spectrum peak means Each delayed scatter cluster. The different cluster with 20 μs and 40 μs delay can be isolated directly. Figure 3 only shows the joint DOA and frequency estimation of the first delay cluster. Each spectrum peak indicates each incoming ray which is included the first delayed cluster with the delay of 20 μs. All of the group and the single ray can be isolated by the temporal filtering process and the spatial smoothing process during the procedure of the proposed algorithm.
4.1. Complexity Analysis
In this part, we hereby make a brief comparison for the algorithmic costs in contrast with the modified MUSIC algorithm and SAGE algorithm. Table 1 gives the comparison of complexity for these algorithms.
When the number of ray estimated is small, it is proved that the SAGE algorithm can achieve coverage within 6–10 iterations through simulation analysis. The complexity of SAGE algorithm is less than that of the modified MUSIC algorithm.
On the contrary, the SAGE algorithm achieves coverage within 20 iterations when the wireless scatter environment is dense multipath. So, the complexity of SAGE algorithm is larger than that of the modified MUSIC algorithm.
From Table 1, it is clearly indicated that modified MUSIC algorithm can be optimal in the dense multipath wireless environment among the methods that we have investigated.
4.2. RMSE Analysis for Specular Model
To compare the estimation performance of both algorithms, we present 300 Monte Carlo’s simulations to assess the angle estimation performance of our algorithm in specular model and define root mean squared error (RMSE) as where is the estimate of DOA of the th Monte Carlo trial. Figure 4 shows the DOA estimation performance comparison with modified MUSIC algorithm and SAGE algorithm in specular models. It is indicated that the performances of angle estimation are nearly the same with modified MUSIC algorithm and SAGE algorithm. However, the modified MUSIC algorithm takes less time than the SAGE algorithm in dense multipath environment.
4.3. RMSE Analysis for Cluster Model
Through the simulation, it is found that the iteration step shows sharp fluctuation for scatter cluster model using the SAGE algorithm. The major reason which leads to the failure of SAGE algorithm in the scatter clusters model is the searching procedure in the M step of SAGE algorithm that does not identify the peak caused by the signals which have same delays.
As mentioned above, the modified MUSIC algorithm is proved to be applied for the cluster models. However, the estimation performance depends on the mean of correlation matrix which substitutes for the expect of the correlation matrix in the practice. So, the number of receive antennas and sample length make a great part in the accuracy of estimation.
Figure 5 illustrates the DOA estimation performance comparison with different elements of receive array. It is clearly shown that the angle estimation performance of our algorithm is gradually improving with the number of antennas increasing. Multiple antennas improve the estimation performance because of diversity gain. Figure 6 shows the DOA estimation performance comparison with different samples. The element of receive array of 8-elements was assumed in this simulation. We confirm that the performance of angles estimation becomes better in collaboration with the increasing samples rate.
As to the conventional SAGE algorithm does not work for these scatter cluster models, in this paper a modification channel parameter method was proposed which combines the temporal filtering techniques and the spatial smoothing techniques based on the conventional MUSIC algorithm. The synthetic channel was then adopted in simulation to evaluate its performance. We come to a conclusion that the modified MUSIC algorithm takes less time than the SAGE algorithm and has less complexity than the SAGE algorithm in dense multipath environment. The results indicated that this method could fulfill parameter estimation with high resolving capability. It is demonstrated that the estimation performance grows with the number of array elements and the sample rate.
This work is supported by the National Science and Technology Major Project of the Ministry of Industry and Information of China (2009ZX03007-003) and the National Natural Science Foundation of China (61101223).
- H. Krim and M. Viberg, “Two decades of array signal processing research: the parametric approach,” IEEE Signal Processing Magazine, vol. 13, no. 4, pp. 67–94, 1996.
- R. O. Schmidt, “Multiple emitter location and signal parameter estimation,” IEEE Transactions on Antennas and Propagation, vol. 34, no. 3, pp. 276–280, 1986.
- R. Roy and T. Kailath, “ESPRIT—estimation of signal parameters via rotational invariance techniques,” IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. 37, no. 7, pp. 984–995, 1989.
- M. Haardt and J. A. Nossek, “Unitary ESPRIT: how to obtain increased estimation accuracy with a reduced computational burden,” IEEE Transactions on Signal Processing, vol. 43, no. 5, pp. 1232–1242, 1995.
- X. Yin, L. Liu, D. K. Nielsen, T. Pedersen, and B. H. Fleury, “A SAGE algorithm for estimation of the direction power spectrum of individual path components,” in Proceedings of the 50th Annual IEEE Global Telecommunications Conference (GLOBECOM '07), pp. 3024–3028, November 2007.
- A. L. Swindlehurst, “Time delay and spatial signature estimation using known asynchronous signals,” IEEE Transactions on Signal Processing, vol. 46, no. 2, pp. 449–462, 1998.
- I. Jaafar, R. Amara, H. Boujemâa, and M. Siala, “Joint angle and delay estimation of point sources,” in Proceedings of the 12th IEEE International Conference on Electronics, Circuits and Systems (ICECS '05), pp. 1–5, December 2005.
- Y. Y. Wang, J. T. Chen, and W. H. Fang, “TST-MUSIC for joint DOA-delay estimation,” IEEE Transactions on Signal Processing, vol. 49, no. 4, pp. 721–729, 2001.
- J. D. Lin, W. H. Fang, Y. Y. Wang, and J. T. Chen, “FSF MUSIC for joint DOA and frequency estimation and its performance analysis,” IEEE Transactions on Signal Processing, vol. 54, no. 12, pp. 4529–4542, 2006.
- 3GPP, “Spatial channel model for multiple input multiple output MIMO simulations,” Technical Specification Group Radio Access Network TR25.996 v6.1.0, 3GPP, September 2003.
- B. Nadler, “Nonparametric detection of signals by information theoretic criteria: performance analysis and an improved estimator,” IEEE Transactions on Signal Processing, vol. 58, no. 5, pp. 2746–2756, 2010.
- J. S. Thompson, P. M. Grant, and B. Mulgrew, “Performance of spatial smoothing algorithms for correlated sources,” IEEE Transactions on Signal Processing, vol. 44, no. 4, pp. 1040–1046, 1996.
Copyright © 2012 Jinsheng Yang et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.