In the last few years, small-cell technology has moved from the realm of academic research to large commercial networks with millions of deployed units. This paper discusses the challenges in deploying and maintaining a large-scale residential small-cell network. The paper first discusses the need for residential small cells to provide ubiquitous coverage in homes located in poor coverage areas of the conventional macrocell, and then describes the architectures of current and next-generation residential small-cell networks, and how these large networks are effectively provisioned and managed. The paper is air interface agnostic and is based on the experience of the authors in commercial deployment of both UMTS and CDMA2000 femtocell networks.

Ever since the first dial-up modems appeared in the 1960s, their obsolescence has been repeatedly predicted. However, contrary to such predictions, dial-up modems thrived in the 1980s and 1990s as a result of the slow rollout of residential digital services and the unprecedented growth of internet and remote access. Since the first 300 b/s dial-up modem standard (V.21), modem speeds have increased steadily. Most recently, International Telecommunications Union (ITU) Recommendation V.90 (1998) takes advantage of the direct digital-network connection of an internet service provider (ISP) remote-access server to achieve speeds of more than 50 kb/s downstream (from ISP to a user). However, for upstream transmission (from a user to ISP), V.90 employs the older V.34 modulation (1994), which typically delivers on the order of 30 kb/s. A new ITU modem standard called V.92 increases upstream rates to above 40 kb/s, again by taking advantage of pulse code modulation connections. In this paper, we present the transmission scheme that has been adopted for V.92. It involves a generalization of Tomlinson-Harashima precoding. We predict that V.92 will be the last dial-up modem standard. However, we have to wonder whether we might be falling into the same trap into which many others have fallen in the past. The future will be the judge!.

A new spectral shaping technique Convolutional Spectral Shaping (CSS), is introduced. In CSS, equivalence classes of possible sign bit sequences are represented by a coset of a convolutional code and a search through the trellis of the code is used to pick a sequence from the selected coset to approximate a target spectral shape for transmission. Simulation results are included to illustrate the performance of CSS. A version of CSS has recently been adopted by the International Telecommunication Union (ITU) in the Pulse Code Modulation (PCM) modem standard V.90.

The minimum mean-squared-error decision-feedback equalizer (MMSE-DFE) has properties that suggest that it is a canonical equalization structure for systems that combine equalization with coded modulation. The structure and performance of the MMSE-DFE are succinctly derived using linear-estimation-theoretic principles in this first part of this two-part paper. The front-end of the MMSE-DFE, called the "mean-square whitened matched filter" (MS-WMF), is preferable in some ways to a matched filter or a whitened matched filter as a canonical receiver front end. In a coded system, the feedback filter of the MMSE-DFE may be implemented in the transmitter using precoding. The MMSE-DFE can perform significantly better than a zero-forcing decision-feedback equalizer, particularly at moderate-to-low SNR's and on severe-ISI channels. The MMSE-DFE is biased. The optimum unbiased MMSE-DFE is the MMSE-DFE with the bias removed. Removing bias improves error probability, but reduces the SNR to SNR/sub MMSE-DFE,U/=SNR/sub MMSE-DFE/-1. It is shown that this SNR relationship is a particular case of a very general result and that SNR/sub MMSE-DFE,U/ gives a more realistic estimate of SNR. The results are extended to partial response equalization and to equalization with correlated inputs in an appendix.>

For pt.I see ibid., vol.43, no.10, p.2582 (1995). The minimum-mean-squared-error decision-feedback equalizer (MMSE-DFE) has properties that suggest that it is a canonical equalization structure in systems that combine equalization with coded modulation. With a given symbol rate 1/T and transmit spectrum, the output signal-to-noise ratio SNR/sub MMSE-DFE,U/ of a MMSE-DFE with an unbiased decision rule is a single parameter that characterizes the channel for coding purposes. Indeed, the transmit spectrum that maximizes SNR/sub MMSE-DFE,U/ is the capacity-achieving (water-pouring) spectrum, and the capacity C(T) (in bits per two dimensions) is given by C(T)=log/sub 2/[1+SNR/sub MMSE-DFE,U/] regardless of the channel characteristics. The performance of a coded system with a MMSE-DFE equalization structure may be accurately estimated using the gain of the coding scheme at a given Pr(E). This performance is shown to be approximately the same as that of a multicarrier system using the same transmit spectrum and similar coding; such systems are known to be able to approach capacity arbitrarily closely. The MMSE-DFE can perform significantly better than a zero-forcing decision-feedback equalizer, particularly at moderate-to-low SNR's and on severe-ISI channels. Simulation results indicate that performance of the MMSE-DFE is surprisingly insensitive to transmit spectral shaping, as long as the transmit spectrum exceeds the capacity-achieving band, but that there is an optimal symbol rate that should (approximately) be used.>

In a two-channel transmission system, the signal-to-noise ratio (SNR) at the receiver can in general be different for the channels. Normally, performance will be dominated by the channel with the worst SNR, but it is possible to do much better. The general scheme proposed provides optimum performance when the SNR varies between the channels and/or when the interference is correlated, without varying the transmitted power for each channel. Simplified schemes attain nearly optimum performance under certain channel conditions.>

High-rate lattice and trellis quantizers for nonuniform sources are introduced and analyzed. The performance of these quantizers is determined by two separable quantities, the granular gain and the boundary gain, which are determined by the shapes of the granular cells and of the support region, respectively. The granular gain and boundary gain are the duals of shaping and coding gain in data transmission applications. Using this duality, it is shown for Gaussian sources that the ultimate achievable boundary gain with high-rate lattice-bounded lattice codebooks is the same as the ultimate gain that can be obtained from variable-rate entropy coding. It is observed that if lattice codebooks can achieve the ultimate granular gain of 0.255 b per dimension, then lattice-bounded lattice codebooks can approach the rate-distortion limit. The performance of lattice quantizers is compared to that of optimum vector quantizers.>

On a linear Gaussian channel with intersymbol interference (ISI), trellis precoding is a method that achieves the equalization performance of Tomlinson-Harashima (TH) precoding, the coding gain of any known lattice-type coset code, and a considerable shaping gain. Trellis precoding may be viewed as a generalization of trellis shaping to Gaussian ISI channels, or, alternatively, as a generalization of TH precoding, with coded modulation that achieves shaping gain. With trellis precoding channel capacity can be approached essentially as closely on any strictly bandlimited, high signal-to-noise ratio Gaussian channel as on the ideal channel, using the same coding techniques. For first- and second-order FIR and IIR (finite and infinite impulse response) channels, it is shown that shaping gains close to 1 dB can be obtained with a two-dimensional four-state trellis code. Trellis precoding is quite practical whenever channel information is available at the transmitter.>

A system is described for high-speed digital subscriber line (HDSL) applications. It uses trellis coding combined with Tomlinson (1971) precoding. Theoretical performance results are provided for uncoded pulse amplitude modulation (PAM) and quadrature amplitude modulation (QAM) assuming an ideal decision feedback equalizer (DFE) that includes an optimum feedforward filter. The similarities and differences in behavior are explained. It is shown that combined coding and Tomlinson precoding can achieve the equalization performance of the DFE and full coding gain of the trellis code. Simulation results are given to show that about 3.4 dB additional margin is obtained by using a four-dimensional eight-state trellis code. The receiver complexity is increased by it is quite manageable.>

The authors investigated the detection of trellis codes designed for channels that are intersymbol interference free when they operate in the presence of intersymbol interference. A well-structured reduced-state sequence estimation (RSSE) algorithm is described which can achieve the performance of maximum-likelihood sequence estimation (MLSE) with drastically reduced complexity. Well-defined reduced-state trellises are first constructed by merging the states of the ML supertrellis using set partitioning principles. Then the Viterbi algorithm is used to search these trellises. A special case of RSSE, called parallel decision-feedback decoding, uses the encoder trellis, yet on channels with large attenuation distortion it can provide a significantly better performance than linear equalization. The performance of RSSE is examined analytically and through simulation, and then compared to that of MLSE and ideal decision-feedback equalization. It is noted that the performance advantage of RSSE can be obtained without significantly increasing the decoding delay or complicating an adaptive implementation.>

On linear bandlimited Gaussian noise channels with sufficiently high SNR, channel capacity can be approached by combining powerful coded modulation schemes designed for Nyquist channels with the equalization power of decision-feedback equalization (DFE). However, this combination may not be realized in a straightforward manner, since, in general, DFE requires delay-free decisions for feedback, and in a coded system such decisions are not sufficiently reliable. A technique is proposed that combines periodic interleaving with noise-predictive DFE, so that delayed reliable decisions can be used for feedback. When sufficient delay in the interleavers can be tolerated, this technique can attain the DFE performance. On severely distorted channels, modest delays can be sufficient to obtain respectable gains over linear equalization.>

A reduced-state sequence estimator for linear intersymbol interference channels is described. The estimator uses a conventional Viterbi algorithm with decision feedback to search a reduced-state subset trellis that is constructed using set-partitioning principles. The complexity of maximum-likelihood sequence estimation (MLSE) due to the length of the channel memory and the size of the signal set is systematically reduced. An error probability analysis shows that a good performance/complexity tradeoff can be obtained. In particular, the results indicate that the required complexity to achieve the performance of MLSE is independent of the size of the signal set for large enough signal sets. Simulation results are provided for two partial-response systems. A simple technique for quadrature partial-response signaling (QPRS) is described that eliminates the quasicatastrophic nature of the ML trellis.>

We consider the convergence and tracking properties of an adaptive multielement integrated receiver structure operating in the presence of impulsive or burst noise. The receiver structure is integrated in the sense that it incorporates both the spatial/temporal predetection processing and demodulation functions in a single entity. This integrated receiver structure has been developed in previous work to approximate maximum-likelihood (ML) reception in the presence of both intersymbol interference (ISI) and spatially distributed interference (SDI) when the latter is modeled as zero-mean Gaussian noise. The present paper is devoted to a simulation study of the convergence properties of this receiver structure when the SDI is replaced by non-Gaussian impulsive or burst noise. Results indicate that the convergence characteristics are relatively insensitive to the statistics of the SDI provided that secondorder properties are held fixed, at least for moderate deviations from Gaussian behavior. For highly impulsive SDI environments, however, the adaptive system exhibits a complete failure to converge.

The structure and performance of a maximum likelihood (ML) receiver for reception in spatially distributed interference channels when a multielement array capability is available are described. Under the assumption of Gaussian interference, the receiver consists of an optimum ML array processor followed by a sequence estimator implemented using the Viterbi Algorithm. Numerical results are provided to illustrate the symbol error probability performance of this optimum ML receiver. A completely adaptive realization of this scheme is described with performance illustrated through simulation.