While the race for a ubiquitous, reliable and cost effective wireless broadband system has generated intense competition, the fundamental problem afflicting the wireless industry is the cost of the transceiver hardware. We describe and discuss the results of some experimental outdoor field trials which demonstrate - for the first time - the hardware feasibility of fixed broadband wireless radio links with spectral efficiencies in excess of 50 bit/s/Hz in severe multipath frequency selective environments (non line-of-sight). Custom cost-effective hardware has allowed data transmission in the range 60-200 Mbit/s occupying a bandwidth of 2-4/5 MHz at MMDS frequencies (/spl sime/2.5 GHz). We also introduce a new modeling methodology inspired by basic theoretical physics which gives a different and practical perspective to the spatio-temporal transmission problem. We exploit the idea that extra-dimensionality can be created by properly sampling the wavefield space created by multipath propagation. Previously reported hardware experiments using the same concept have addressed the simpler problem of indoor communications in a 30 kHz cellular channel. The developed technology has the potential of making practical and cost effective the deployment of high capacity microwave data communications equipment for point to point and point to multipoint systems.

Traditional multicarrier techniques perform a frequency-domain decomposition of a channel characterized by frequency-selective distortion in a plurality of subchannels that are affected by frequency flat distortion. The distortion in each independent subchannel can then be easily compensated by simple gain and phase adjustments. Typically, digital Fourier transform schemes make the implementation of the multicarrier system feasible and attractive with respect to single-carrier systems. However, when the channel is time-frequency-selective, as it usually happens in the rapidly fading wireless channel, this traditional methodology fails. Since the channel frequency response is rapidly time-varying, the optimal transmission/reception methodology should be able to process nonstationary signals. In other words, the subchannel carrier frequencies should be time-varying and ideally decompose the frequency distortion of the channel perfectly at any instant in time. However, this ideally optimal approach presents significant challenges both in terms of conceptual and computational complexity. The idea disclosed in this work is that a nonstationary approach can be approximated using signal bases that are especially suited for the analysis/synthesis of nonstationary signals. We propose in fact the use of a multicarrier system that employs orthogonal signal bases of the chirp type that in practice correspond to the fractional Fourier transform signal basis. The significance of the methodology relies on the important practical consideration that analysis/synthesis methods of the fractional Fourier type can be implemented with a complexity that is equivalent to the traditional fast Fourier transform.

The problem of detecting a sequence using the maximum-likelihood strategy, when the time-varying multipath channel is unknown, can be in principle solved using the generalized likelihood detector (GLD). The GLD metric involves finding the orthogonal projection of the received signal onto the subspace of the transmitted signal distorted by the multipath channel: the faded signal corresponding to one of the transmitted sequences is in fact known to lie in a subspace but its exact location is not known, because the channel parameters are unknown. This detector (which in the case of a static channel collapses to the per-survivor processing method), is also called a matched subspace detector because its statistic is "matched" to the a priori known signal subspace. Unfortunately, the computation of the (perfectly matched) orthogonal projection of the received signal onto the multipath faded signal subspace is, in the general time-varying case, impossible. We introduce in this work the idea of using new wavelet-based subspaces that approximate the original signal subspace. The nested sequence of linear vector spaces, defined by a wavelet-based multiresolution decomposition of the fading channel time variations, provides a set of subspaces that, at an increasingly high level of detail, are "efficient" representations of the original signal subspace. For each of these representations sequence detection at different levels of resolution can be performed.

The creation of multiple channels (sharing the same time-frequency region) between a transmitter and a receiver can be achieved by sampling the wavefield space with respect to the spatial domain. The wavefield space is the signal space spanned by the channel parameters which characterize the multipath fading environment. Using this basic intuition it is possible to reliably transmit data at rates exceeding the famous Shannonian spectral efficiency of Log/sub 2/(1+SNR) bit/sec/Hz where the SNR is the signal-to-noise ratio over the bandwidth occupied by the signal. This exciting possibility has generated significant interest in the communications community. However, it is not, clear at this point how a realistic and cost-effective radio could take advantage of such an enhancement. WJ Communications has invested significantly in the research of this subject and sponsored the study of a new transmission scheme, STREAM/sub TM/, [Spatial Transmission with Radio Enhanced Adaptive Modulation] able to achieve very high speed bandwidth-efficient wireless data communications in arbitrary environment: outdoor/indoor, mobile/fixed, line-of-sight/obstructed. We describe here the general ideas of the technology and show the results of the first hardware prototype which has no counterpart currently in industry in terms of scaleability, cost, performance.

Multicarrier systems of the OFDM-type perform a frequency domain decomposition of a channel characterized by frequency selective distortion in a multitude of subchannels that are affected by frequency flat distortion. The distortion in each independent subchannel can then be compensated by simple gain and phase adjustments, somehow overcoming the need of one of the most complex components of a digital modem: the equalizer. On the other hand, coding and transmission power assignments can be applied across the subchannels in a way that resembles the Shannon water pouring argument. As the bandwidth of the subchannels is made sufficiently small one can asymptotically reach the channel capacity. We prove that a careful implementation of the analog RF front-end is extremely critical to the asymptotic behavior of multicarrier transmission in approaching capacity. In fact multicarrier signals exhibit extreme vulnerability to distortions induced by RF electronics. We concentrate on the characterization of those degradations that can not be compensated at baseband by means of digital signal processing and show that the throughput achievable by realistic OFDM modems is much lower than what typically advertised, particularly in transceivers with low performance analog RF front-end.

New methods for maximum-likelihood sequence estimation based on the Viterbi algorithm (VA) are presented. In the proposed scheme, the channel estimator and the Viterbi processor operate concurrently. At any given time-step, the sequence provided to the channel estimator comes from the survivor with the best metric value. These already known modifications of the traditional decision-directed VA cause large variance in the estimated channel coefficients. In fact, sequences with a high error rate may be used to perform estimation, and also the adjustment term of the channel tracking algorithm may exhibit abrupt changes caused by a "survivor swap", (that is by the event in which a different survivor has the best metric at step n, with respect to the step n-1). The proposed regularization procedure forces the channel vector to lie in the appropriate a priori known subspace: while the variance decreases, a certain amount of bias is introduced. The variance-bias tradeoff is then automatically adjusted by means of a cross-validation "shrinkage" estimator, which is at the same time optimal in a "small sample" predictive sum of squares sense and asymptotically model mean squared-error optimal. The method is shown by means of hardware experiments on a wide-band base station to be extremely more effective than per survivor processing, minimum survivor processing, and traditional decision directed approaches.

We propose new detectors for direct-sequence code-division multiple-access (CDMA) signals that outperform known approaches in rapidly fading multipath channels. Multipath compensation in CDMA systems is a problem of significant complexity especially when rapidly fading is afflicting the radio frequency channel. In this work, we depart from typical approaches in search of new kernels that can more accurately characterize the time-varying nature of the estimation problem and focus on a multiresolution representation of the fading processes. The unknown channel time variations are in fact decomposed using optimal unconditional bases in the family of the orthonormal wavelets. We show that it is possible to represent the channel in a reduced-order dimensional space by matching the scattering function of the multipath channel to its decomposition and obtain an approach that is effective in fast fading environments, such as those practically found in macrocell wireless communication applications. We apply this representation to the development of a practical multiscale filter that achieves multiuser separation minimizing a time-averaged squared error. The technique is studied by means of computer simulations and hardware experiments that employ a currently deployed base-station system.

A new approach based on multiscale decomposition and higher-order statistics is presented for the simultaneous solution of multiuser interference and time-varying multipath propagation in the uplink of a cellular direct-sequence spread-spectrum code-division multiple-access (DS/CDMA) system. Each channel between the mobile transmitter and the base-station receiver is unknown and arbitrarily varying with time. The optimum filter achieving separation and multipath compensation is time-variant. The typical approach in many multiuser detectors previously proposed is to assume that the channel is almost static (time invariant) and attempt detection according to this model. Slow variations of the channels are then compensated using adaptive algorithms that force the estimates (of the channels or of the separating filters) to be constantly in search of a convergence point. If the channel coefficients variations in time are fast with respect to the convergence time of the adaptive algorithm, significant degradation may result. In this work, we depart from this traditional approach and we investigate new kernels that more accurately can characterize the time-varying nature of the detection problem. As a first step, we show that the super-exponential framework can still be applied in a time-variant environment. Then, we describe a multiresolution decomposition of the filter components, essentially constraining its variations in time to remain within the solution subspace. The resulting algorithm overcomes some of the drawbacks associated with slow convergence and insufficient tracking capability typical of many blind approaches and several nonblind methods based on the slow fading assumption.

The antenna array processing problem in the reverse link of the current US digital cellular communication system is studied and higher than second-order statistics (HOS) baseband processing is proposed as a possible candidate solution. The approach, based on the minimization of the mean squared error using fourth-order cumulants alone, is non-blind because the knowledge of the synchronization sequences of the IS-136 system is fully exploited. A recursive total least squares algorithm is used in the adaptive implementation to mitigate the effects of high error variance in the estimates of the cumulants based on sample-statistics. The method is very effective in a fast fading environment and strong cochannel interference.

The channel tracking problem in the reverse link of a TDMA cellular system exploiting cyclostationary higher order statistics (CHOS) is studied. An adaptive algorithm based on fourth-order cyclic cumulants of the fractionally spaced baseband received signal is derived. The algorithm reveals from experimental verification improved performance with respect to the traditional MMSE method.

A new solution to blind beamforming for signals that exhibit higher-order cyclostationarity is presented. Exploiting some previous theoretical developments, we show how cyclic cumulants of the received signals can be used to obtain the weights of the beamformer that performs blind extraction. The method is based on a spatial interpretation of a deconvolution procedure known as the superexponential algorithm. The basic block processing algorithm is made fully adaptive using an adaptive total least squares scheme and applied to a typical mobile communications scenario where several cochannel interferers corrupt the signals of interest.

The antenna array processing problem in the reverse link of the current US digital cellular communication system is studied and higher-than-second-order-statistics (HOS) baseband processing is proposed as a possible candidate solution. The remarkable difference of our approach as compared to other existing similar techniques is the idea of the minimization of the mean squared error using fourth-order cumulants alone and nonblind criteria. A recursive Jacobi total least squares algorithm is used in the adaptive implementation to mitigate the effects of high error variance in the estimates of the cumulants based on sample statistics. The method is shown to be very effective in a fast fading environment with multiple cochannel interferers.

We show that the popular subspace method for identification of multichannel systems can be implemented using particular contractions of higher-order cumulants, in the case of independent, identically distributed (i.i.d.) inputs. The algorithm appears to outperform the traditional correlation-based subspace algorithm in low signal-to-noise-ratio (SNR) and/or correlated noise.

A new adaptive algorithm for blind interference rejection and multipath mitigation is studied and applied to antenna array processing in the reverse channel of a time-division multiple-access (TDMA) cellular communication system. The method is based on higher order statistics (HOS) processing of the baseband vector samples at the antenna array output. The similarity between the cumulant-based solution and the standard multivariable least-squares solution is exploited to derive an efficient adaptive algorithm based on a low-rank processing architecture. The algorithm exhibits good tracking and enhanced identification capability with respect to traditional least-squares methods.

A new update algorithm for space-time equalization of wireless time-division multiple access signals is presented. The method is based on a modified QR factorization that reduces the computational complexity of the traditional QR-decomposition based recursive least squares method and maintains numerical stability. Square roots operations are avoided due to the use of an approximately orthogonal transformation, defined complex scaled tangent rotation.

A new adaptive algorithm for blind interference rejection and multipath mitigation is studied and applied to antenna array processing in TDMA cellular communication systems. It is shown how the estimation of multiple signals from different sources by means of a multi-sensor receiver can be formulated as a multi-channel deconvolution problem. The method proposed is based on high-order statistics (HOS) processing of the baseband vector samples at the antenna array output. The similarity between the cumulant-based solution and the standard multi-variable least squares solution is exploited to derive an efficient adaptive algorithm based on the vector lattice architecture. The algorithm is numerically stable, considerably less complex than other existing multi-channel methods using HOS processing and exhibits rapid convergence with respect to blind array processing algorithms using simple gradient-based minimization procedures.

A linear high-order statistics based method is proposed to mitigate the performance degradation caused by multipath RF propagation in a mobile radio communication system using a linear antenna array in the base station receiver. The method is based on noncausal autoregressive modeling. It is shown that a set of linear equations (involving only cumulants of the received baseband digitized signal) can be obtained to perform noniterative deconvolution. An efficient adaptive algorithm is proposed.