FBMC-PAM is a recently introduced filter bank multicarrier technique which exploits pulse amplitude modulation and operates at the same symbol rate as OFDM, unlike FBMC-OQAM which operates at twice that symbol rate. The scheme is efficiently implemented in the frequency domain and, in the system, the receiver is equipped with a frequency domain equalizer. In the present paper, uplink access is considered, focusing on two issues, single carrier modulation and preamble-based channel estimation. It is shown that the scheme can handle asynchronous users, thanks to the spectral separation brought by the filter bank and flexible frequency domain equalizers, and the performance is demonstrated by simulations. An FBMC-PAM-based physical layer can be envisaged for the uplink of future wireless systems.

In spite of widely recognized benefits, the acceptance of filter bank multicarrier (FBMC) as an alternative to cyclic prefix orthogonal frequency division multiplexing (CP-OFDM) for the physical layer of future mobile and wireless communication systems is restrained by the introduction of offset-quadrature amplitude modulation (OQAM). The proposed approach, based on the so-called lapped transform, combines a simple yet efficient perfect reconstruction real filter bank with QAM modulation. Real processing of the real and imaginary components of the QAM data samples introduces a symmetry of the used carriers and the spectrum of the multicarrier signal can be either continuous or fragmented. Implementing the receiver with an extended FFT allows for frequency domain channel equalization and the performance achieved is sufficient to avoid the need for a cyclic prefix. The scheme is appropriate for asynchronous access and the filter provides a high level of spectral protection, as required by users in cognitive radio scenarios.

The paper is devoted to the performance analysis of a new receiver structure recently introduced for OFDM/OQAM systems. Such a structure is based on the frequency-despreading at the receiver and it is designed to minimize the processing delay of the receiver structure. Its performance analysis and the comparison with the performance of the standard receiver structure show the capability of the new structure to operate with a small number of subcarriers.

The interpolation of complex signals is frequently encountered in the signal processing chains of high performance transmission systems. Linear interpolation might be inappropriate for this kind of situation and, here, geometric interpolation is proposed instead. Specifically, it is proposed, in a Filter Bank-based Multi-Carrier/Offset QAM (FBMC/OQAM) transmission system, to use geometric interpolation to determine the frequency domain intermediate points of each sub-channel needed to calculate the T/2-spaced sub-channel equalizer coefficients at the beginning of the burst. The equalizer coefficients are computed using the inverse fast Fourier transform (IFFT). This equalizer must compensate ISI and ICI caused by a selective channel and timing offset. Simulation results demonstrate the effectiveness of the proposed equalizer.

The filter bank multicarrier (FBMC) technique has been proposed in the FP7 European research program, as an enhancement to the OFDM technique for future broadband radio systems. Besides gains in efficiency, mainly due to the absence of cyclic prefix, an important feature of FBMC is the absence of spectral leakage, which makes it particularly appealing for cognitive radio. However, in burst transmission, the filter impulse response introduces transitions which extend the length of the burst with respect to OFDM and may reduce the gains in efficiency. After a brief overview of the FBMC technique, the exploitation in burst transmission is discussed. The effect of cutting the initial and final transitions of the burst is investigated. The level of temporary spectral leakage introduced by this operation is assessed. It is shown that cuts leading to an increase in burst length of half a symbol with respect to OFDM affect the first and the last symbol only, resulting in a minor reduction in bit loading for these symbols. Application to the time division duplex access mode is considered, as well as the case of the presence of a preamble in the burst. For implementation, it is proposed to resort to a simple memory preloading technique, which has the advantage of compatibility with OFDM in dual mode systems.

A global assessment of the performance of gradient descent algorithms based on the constant modulus criteria CM(2,2) and CM(1,2), and denoted CMA2-2 and CMA1-2 respectively, is provided. To begin with, it is shown that no theoretical excess error with respect to the MMSE solution is introduced by the CM(2,2) criterion if the source signal satisfies the usual conditions and its kurtosis is 4/3, and the excess error introduced by the criterion CM(1,2) is negligible. Then, existing gradient excess mean-squared error estimations are confirmed through a different approach and they are completed by time constant estimations. Finally, for the same minimum mean-squared error, CMA2-2 is shown to be twice as fast as CMA1-2, which confirms the superiority of the CM(2,2) criterion in equalization with non-CM signals.

The MMSE criterion introduces a bias of the coefficients in adaptive filtering. For constant modulus signals, an estimation of this bias is derived. Constant modulus (CM) criteria lead to different values of this bias. A simple approach is presented to derive estimations for the two criteria of practical interest, namely CM(1,2) and CM(2,2), and it is shown that the bias with CM(1,2) is smaller than the bias with MMSE, while the bias with CM(2,2) is larger. The case of non constant modulus signals is also considered. The results presented are likely to make the performance of CM algorithms better understood and they should contribute to stimulate their introduction in practical systems

The two most important constant modulus criteria are studied and compared, exploiting recently obtained results. A theoretical analysis of the performance is provided and excess output MSE figures are derived. The answer to the title question is found to depend on the output error power. In applications where the lower bound for the output signal-to-noise ratio is small, typically less than 8 dB, the CM(2,2) criterion can be employed. Otherwise, the CM(1,2) criterion is preferable. This result can be of great help to system designers, to select the constant modulus criterion that best suits their application and their performance objectives.

The paper presents a statistical analysis of the multi-split LMS algorithm. Deterministic recursions are obtained for the mean weight vector and the mean square error. Simulation results display excellent agreement with the theoretical predictions, and enable us to validate the proposed models for both transient and steady-state behaviors.

This paper presents an analysis that justifies the improved performance of the multi-split LMS algorithm. It is shown that instead of reducing the eigenvalue ratio, the multi-split operation increases the diagonalization factor of the transformed input signal autocorrelation matrix, which assists the power normalized and time-varying step-size LMS algorithm used for updating the single parameters independently. Case studies and simulation results enable us to evaluate the improved performance of the multi-split LMS algorithm.

A spatial domain interpolation technique has been proposed recently for the high-quality restoration of missing blocks in images . An efficient implementation is presented here, which leads to a little more than one multiplication per calculated pixel. Moreover, it is shown that the multiplications can be eliminated altogether, if a slight loss in quality can be tolerated. The technique can be applied in postprocessing devices in image communication terminals.

The transmission of high bit rates over power lines can be severely limited by large recurrent spurious sinusoidal impulses, even if a robust multi carrier scheme is employed. To cope with this impairment, it is proposed to attenuate the impulse in the received sequence through signal processing techniques. The occurrence of the sinusoidal impulse is, first, detected and its samples are identified. Then, the corrupting samples are reconstructed from parameter estimations and they are subtracted from the received sequence. The simulation results, obtained with a multi carrier scheme based on filter banks and operating in the CENELEC band A, are presented and discussed. They show that a scheme with moderate complexity can attenuate the disturbing impulse sufficiently to restore most of the capacity of the transmission system.

This paper presents the implementation of blind equalizers with a new filtering structure, the split transversal filter. The aim is to improve the robustness of known blind algorithms. Comparisons between the new scheme and the traditional structure for three of the Bussgang techniques are presented. While the traditional equalizers are attracted by a local minimum, the split equalizer can avoid them and reach the global solution.

The idea of multi-split adaptive filtering is applied to the DFE. The feedforward and feedback sections in the DFE are divided into parallel sub-filters by imposing separately symmetry and antisymmetry conditions on the impulse responses of the filters by using appropriate and distinct sets of linear constraints. The purpose of this is to obtain a new structure that gives better performance than the conventional structure. The performance of multi-split DFE is tested with different algorithms (LMS, RLS, DCT-LMS) and compared to that of a multi-split linear transversal equalizer.

The specifications of filter banks for multicarrier transmission systems with a large number of subchannels are discussed, with application to xDSL and power line communication in mind. The near perfect reconstruction (PR) modulated approach is considered and the importance, for the system, of the prototype filter delay is stressed. An existing design technique known to be particularly relevant to the context is revisited from a frequency sampling perspective. The performance results in terms of subchannel noise floor and delay are given for several filter lengths and an experimental validation is provided. Finally, an improvement to the design technique is proposed, which brings a gain of 3.3 dB in subchannel interference power level.

A novel transversal split filter configuration is proposed and the split optimum Wiener filter is introduced, as well as the symmetric and antisymmetric linear phase Wiener filter. The approach consists of combining the idea of split filtering with a linearly constrained optimization scheme. Then, the continuously split procedure is introduced and the multi-split adaptive filter is derived. It is also shown that such a structure can be viewed as a Hadamard-domain adaptive filter when the number of coefficients is set to a power of two. Simulation results obtained with the normalized LMS algorithm are presented and compared with DCT-domain adaptive filter.

We introduce the optimum interpolated Wiener filter, which resulted from an original and elegant linearly-constrained approach to the interpolated transversal filtering problem. An accurate analytical model for the mean coefficient behavior of LMS-AIFIR filters is also presented. Simulation results illustrate the effectiveness of the proposed model.

A new structure for adaptive interpolated finite impulse response (AFIR) filtering is presented, in which the RLS algorithm can be directly applied for updating only the nonzero coefficients of the sparse filter. The approach is based on the linearly-constrained filter called the generalized sidelobe canceller. The performance of the RLS algorithm applied to the AIFIR filtering is illustrated by simulations and compared with that of the LMS algorithm.

An efficient method is presented to restore a block in an image, using a reduced set of border pixels. The basic assumption is that, if the set of pixels consisting of the missing block and its border pixels is transformed by the DCT, the high frequency coefficients obtained can be set to zero. A system of linear equations results, whose solution yields approximations for the values of the lost pixels. In this interpolation scheme, only eight border pixels are used in the computation of any missing pixel. Simulation results show that the approach generally outperforms alternative methods at a smaller computational cost.

This paper compares the eigenstructure and modulation algorithms, which are used for two-channel lossless FIR filter optimization. We study the effects of eigenvalue separation of the input covariance matrix and the step size on their convergence behavior. First, we show that the convergence rate of two algorithms increases as the separation of eigenvalues of the covariance matrix increases. The modulation algorithm converges more rapidly than the eigenstructure one because of its better eigenvalue separation. Second, the necessary condition for which the two algorithms converge is derived. Simulations are presented which support the analysis.

In multicarrier data transmission using filter banks, adaptive equalizers can be introduced in the receiver in every sub-channel, to achieve high bit rates. Following conventional data transmission techniques, two approaches can be envisaged, namely the double sampling equalizer (DSE) and the critical sampling equalizer (CSE). Both schemes are discussed and assessed, in the multicarrier context. Estimations are given for the lengths of the equalizers as a function of the channel distortion and of the roll-off factor of the prototype filter in the receiver filter bank. Simulation results associated with two channel models are given to support the theoretical analysis.

A novel computationally highly efficient realization of sharp symmetrical bandstop FIR filter is proposed. The new structure is derived using the frequency-response-masking technique, where the bandedge-shaping filter is derived from a half-band filter by substituting each delay of the half-band filter by M delays. The masking filters are unconventional. They are quadrature filters derived from linear combinations of the masking filters in the conventional frequency-response-masking technique. Approximate expressions for the optimal value of M and the corresponding number of multipliers are derived.

In many processing and coding techniques, images are decomposed into blocks of pixels and the DCT is used to transform those blocks. In some situations, blocks are dropped or lost, and, at some point in the processing, they must be restored from their neighborhood in the images. It is shown that an exact interpolation of image blocks can be performed in the spatial domain, using the DCT. The conditions for a solution to exist are derived and the corresponding algorithm requires 8 multiplies per restored pixel. The exact interpolation approach outperforms the alternative methods in the simulations presented, at a significantly smaller computation cost.

Frequency specifications imposing small transitions between pass-bands and stop bands lead to FIR filters having long impulse responses and highly correlated coefficients. Multirate techniques can reduce the resulting redundancy and the combination of a sharp undersampled filter with a couple of interpolating filters can lead to significant gains in arithmetic complexity. The alternative implementation proposed, and the corresponding filter design can bring an additional gain in the total operation count.

A robust approach to implement the multidimensional fast least squares (FLS) algorithm for linearly-constrained adaptive spatial/temporal filtering is presented in this work. It can be directly applied in adaptive beamforming for broadband signals. Simulation results illustrate the performance of the proposed approach which is compared with the classical LMS based Frost algorithm.

A robust approach to implement the FLS algorithm for linearly constrained adaptive filtering is derived in this work. The robustness is provided by means of an additional correcting term which is also updated by a LS procedure. In fact, the novel algorithm works as the LS version of the classical LMS-based Frost algorithm. Simulation results with a long data input sequence show the performance of the proposed technique.>

An important international standardization effort has been made for still image compression. Two standards have been elaborated, namely, the IS 10918 or more commonly the JPEG (Joint Photographic image Expert Group) for continuous-tone images and the IS 11544 or JBIG (Joint Bi-level Image expert Group) for black-white images. Though JPEG has a lossless operation mode, it addresses mainly lossy compression. We report that some gain in the lossless compression ratio can be obtained by representing the original image with a Gray code, separating the bit plane and applying the JBIG standard. A gain of 15% to 22% is achieved by the Gray code with respect to the usual binary code. Some results obtained with a general data compression algorithm, namely, the GZIP (which is based on an Lempel, Ziv, Storer and Szymanski method cascaded with an adaptive Huffman coding) combined with Gray coding and bit plane separation are also presented.>

Radio receivers for digital cellular systems must respect tight constraints in terms of selectivity and dynamic range. Delta-sigma analog-to-digital converters offer a response to the last problem thanks to their high resolution. On the other hand, selectivity is often achieved by filtering the received signal at an intermediate frequency (IF). Limitations of A/D converters impose usually an extra IF conversion of the signal to a lower frequency before being able to properly digitize it. We propose a new delta-sigma converter operating directly on the IF signal allowing the design of simple receivers offering both a high resolution and a good selectivity. Undersampling the IF signal allows reduction of the load of the digital section of the receiver.>

A fast least squares algorithm is proposed for adaptive filtering with linear constraints. The algorithm is exact, with a computational complexity proportional to NK+K/sup 2/+N+K where N is the number of the filter coefficients and K is the number of constraints. The approach consists of obtaining the adaptation gain by means of a fast least squares (FLS) algorithm and using this gain in the recursive calculations of the parameters related to the constraints. The recursive procedures are carried out by using the matrix inversion lemma. The good performance of the proposed method is illustrated by some simulation results. The algorithm can be used in several applications such as array processing, adaptive beamforming, spectral analysis and telecommunications.>

ADAPTIVE is a highly interactive and visually oriented software package for the design, analysis, and simulation of adaptive filters. Software packages such as ADAPTIVE can be used to introduce new users to adaptive filtering and develop their skills in a manner not possible with text materials alone. ADAPTIVE will also be shown to be useful to the experienced user. It is anticipated that adaptive filtering can become accessible and familiar through the use of computer-aided-engineering (CAE) tools, much in the same way that digital signal processing (DSP) has found its way into the mainstream.>

A generic representation of fast least squares algorithms based on the QR decomposition technique is given, pointing out the equivalence with the normalized lattice. The numerical stability observed in simulations is justified, and the implementation in adaptive filters is discussed, with emphasis on the computational complexity issue.>

The fact that lattice and QR methods of adaptive least-squares filtering follow from the same geometric framework allows the first solution to the parameter identification problem using fast QR techniques. These relations suggest combining QR and lattice algorithms into hybrid algorithms of low complexity and good numerical behavior. It is emphasized that many other possibilities are available for mixing and matching the steps of QR and lattice algorithms. The various algorithms so obtained differ in their computational complexity, accessibility to desired variables, etc., but all work within a related geometric framework.>

The signal parameter vector that is computed and updated in the fast QR adaptive filter can be related to prediction and signal model coefficients, and several results concerning these relations are presented. In particular, these results provide information about the signal model, whether it is a moving-average or a predictable signal, as well as about the model order. The results can be useful in real-time signal analysis and can make the QR approach more attractive in adaptive filtering applications.>

An adaptive notch filter based on a fast recursive least squares (FLS) algorithm is studied. The adaptive procedure takes place in the FIR (finite impulse response) section of the filter and the IIR (infinite impulse response) section is updated each time with the most recent coefficients of the other section. The implementation can take the form of a cascade of second order sections. Overall, the FLS notch filter has about the same complexity as the transversal adaptive filter, but performs better for the analysis of narrow-band signals in high-level noise.>

The introduction of Fast Least Squares (FLS) algorithms in real transversal adaptive filters implies that some modifications be made to the theoretical algorithms, that engineering parameters be introduced, and that information be provided to guide the practitioner in the selection of parameter values. These aspects are discussed in the present contribution. Implementation on a programmable signal microprocessor confirms, in broad lines, the theoretical results.

The evolution of telecommunications towards an Integrated Services Digital Network (ISDN) offers new opportunities for signal processing applications. Recent progress in basic techniques, like perfect signal decomposition and reconstruction or Least Squares adaptive filtering, are crucial in that evolution. Beyond the emergence of new equipment, signal processing advances are paying the way for artificial intelligence techniques, which will accompany the "information age" promised by the ISDN.

A stability condition analysis is performed for weighted Least Squares Algorithms. It is shown that the introduction of a constant in the prediction error energy updating recursion can make the algorithm stable. Coefficient wordlength estimations can then be derived from those available for Least Mean Squares Algorithms. The results given are useful in designing adaptive filters based on least squares techniques in their simplest forms.

A simple exponential technique is presented for the real time power estimation of non stationary signals. It is used to adjust the coefficient adaptation step size of a LMS filter. Its efficiency is demonstrated with an adaptive linear predictor for speech. Improved performance with respect to conventional techniques appears. However the main advantage of that exponential estimation is its simplicity, since it can be implemented with just one, or even no full multiplications or divisions.

Leakage terms are introduced in certain adaptive filters to improve stability and robustness to transmission errors. The effects of that modification to the conventional procedure is analyzed for the gradient technique. Estimations are given for the resulting bias of the coefficients and error signal power increase. As an illustration, a prediction filter in an ADPCM codec is considered.

The field of applications of adaptive digital filters is growing and, as for fixed coefficient filters, system designers and project engineers need simple means for computational complexity assessment. An attempt is made in the present paper to derive simplified expressions for coefficient and internal data wordlengths as a function of specified parameters. Only the gradient technique in direct and cascade forms is considered.

The quality standards recommended by the CCITT are the reference for the design of digital transmultiplexer filters. Their impact on filter specifications is analyzed, and expressions are given for the filter order, coefficient, and data word lengths in FIR and IIR approaches. Improvements on the polyphase filter bank design are pointed out, and the method is presented as a hybrid FIR-IIR technique. Finally, it turns out that the multiplication rate is the crucial parameter for complexity comparison between the approaches considered.

A 32 kbit/s rate per channel would double the capacity of digital links compared with 62 kbit/s standard PCM and permit digital techniques to be used throughout in telecommunication networks. The requirements imposed on a coding scheme for long-haul telephone transmission are reviewed. Besides the minimal complexity of implementation, the toll-quality level, the capability of handling signals other than speech, a crucial feature is the non cumulation of degradations in successive coding-decoding processes. An ADPCM coder is proposed ; in the implementation with finite precision arithmetic, both adapted quantization step sizes and prediction filter outputs track in coder and decoder. Experimental results show that the 32 kbit/s scheme outperforms 56 kbit/s PCM for speech in several aspects and is adequate for data transmission up to 4800 bit/s.

The multiplexing scheme presented here applies to multirate filtering with half-band FIR filters which must be implemented with the help of a single multiplier. It consists of interleaving the data samples of the filter stages, according to the binary rate multiplier principle. For the realization of the stages, the transposed direct form is retained with serial processing of the data. No extra buffer memories are requised between stages and control signals are easily generated.

Techniques applied to the digital processing of speech and signaling in a 60 channel FDM to 30 channel PCM converter are presented. Great efficiency is achieved through connecting a particular type of Fourier Transform computer to a polyphase network in which coefficient symmetries are exploited. Digital processing parameter values have been chosen so as to meet specified performance objectives. Realization of a laboratory model is outlined and measured performance is reported. Adequacy of the various options retained for the digital processing is confirmed by the experimental results.

The cascading of a discrete Fourier transform processor and a digital polyphase network is shown to reduce the computation rate in frequency multiplexing-demultiplexing systems to a value close to minimum. Implementation advantages of that technique are pointed out. The highest efficiency is achieved when the number of channels is close to a power of 2, which is demonstrated by the 60-channel frequency-division-multiplexing (FDM)-2 × 30-channel time-division-multiplexing (TDM) transmultiplexer; a rough estimate of the computation rate is carried out in a practical case and appears to be quite within reach of the present technological capabilities. Significant cost advantage over equivalent analog equipment is expected. A digital version of the 12-channel FDM system using the same technique is also considered.