Training large language models (LLMs) with limited computing resources is challenging because of their immense memory space requirements. In this paper, we specifically focus on the scenarios where we have insufficient aggregate GPU memory to store all model states but explore pipeline parallelism and offloading across all system resources to train the model. In this context, SPipe presents a hybrid GPU and CPU pipelining mechanism that consists of two pipelines: a GPU pipeline to reduce the bubbles in conventional pipeline parallelism and a GPU-CPU pipeline to alleviate data transfer overhead and CPU bottlenecks in offloading data and computation. We evaluate SPipe for training LLMs of various sizes with diverse configurations in practice. The result indicates that SPipe outperforms the state-of-the-art by $1.26 \times$.

Weakly-supervised semantic segmentation aims to train a semantic segmentation network using weak labels. Among weak labels, image-level label has been the most popular choice due to its simplicity. However, since image-level labels lack accurate object region information, additional modules such as saliency detector have been exploited in weakly supervised semantic segmentation, which requires pixel-level label for training. In this paper, we explore a self-supervised vision transformer to mitigate the heavy efforts on generation of pixel-level annotations. By exploiting the features obtained from self-supervised vision transformer, our superpixel discovery method finds out the semantic-aware superpixels based on the feature similarity in an unsupervised manner. Once we obtain the superpixels, we train the semantic segmentation network using superpixel-guided seeded region growing method. Despite its simplicity, our approach achieves the competitive result with the state-of-the-arts on PASCAL VOC 2012 and MS-COCO 2014 semantic segmentation datasets for weakly supervised semantic segmentation. Our code is available at https://github.com/st17kim/semantic-aware-superpixel.

In this paper, we consider a beamforming vector design and device selection problem in over-the-air computation (AirComp) for federated learning. Since the learning performance improves as more devices participate in the federated learning aggregation, we formulate a beamforming vector optimization problem that maximizes the number of selected devices under a given target aggregation mean-squared error. This AirComp uplink beamforming problem with device selection is shown to have the same form as the downlink multicast beamforming problem with user selection, which establishes the AirComp-multicasting duality. We design a low-complexity algorithm based on the projected subgradient method that is orders of magnitude faster than conventional semidefinite relaxation-based algorithms and faster than local model training on the devices, which makes it possible to implement the proposed wireless federated learning in real time. Numerical results show that the proposed algorithm provides significant multiple antenna beamforming gains and achieves the performance of the ideal federated learning system with no aggregation errors.

Since the state-of-the-art quantum computers are still noisy and error-prone, classical simulation of quantum circuits is essential in verifying/calibrating quantum computers and prototyping/debugging complex quantum algorithms. Classical simulation of large quantum systems is challenging due to its exponential increase in space and computation requirements. In this paper, we propose a full-state simulation framework, SnuQS. It exploits storage devices, such as HDDs and NVMe SSDs, to enlarge the available main memory capacity at a small cost. To achieve maximum I/O bandwidth, we propose an overlay-based memory management technique and optimization techniques. We also propose an I/O subsystem architecture that guarantees the maximum bandwidth of each storage device. We evaluate SnuQS on a 64-core CPU and 4-GPU system with 80 2TB HDDs and 10 4TB NVMe SSDs using quantum supremacy and quantum Fourier transform circuits. The experimental result indicates that SnuQS and the proposed I/O subsystem together is an effective and practical solution to scale the full-state simulation of large quantum circuits at about 300X lower cost than the DDR4 DRAM main-memory-only system.

This paper proposes a framework called SnuRHAC, which provides an illusion of a single GPU for the multiple GPUs in a cluster. Under SnuRHAC, a CUDA program designed to use a single GPU can utilize multiple GPUs in a cluster without any source code modification. SnuRHAC automatically distributes workload to multiple GPUs in a cluster and manages data across the nodes. To manage data efficiently, SnuRHAC extends CUDA Unified Memory and exploits its page fault mechanism. We also propose two prefetching techniques to fully exploit UM and to maximize performance. Static prefetching allows SnuRHAC to prefetch data by statically analyzing CUDA kernels. Dynamic prefetching complements static prefetching. SnuRHAC enforces an application to run on a single GPU if it is not suitable for multiple GPUs. We evaluate the performance of SnuRHAC using 18 benchmark applications from various sources. The evaluation result shows that while SnuRHAC significantly improves ease-of-programming, it shows scalable performance for the cluster environment depending on the application characteristics.

In this article, we present a new iterative MIMO detection algorithm based on inexact alternating direction method of multipliers. Each iteration is considered as a neural network layer with learnable parameters, which are optimized by the stochastic gradient descent algorithm with a training data set of the received vectors and the ground truth transmitted signals. Numerical results show that the proposed algorithm outperforms the existing learnable detection network and it achieves near-optimal performance close to the sphere decoder in the case of a large number of receive antennas.

In this paper, we propose a runtime, called HUM, which hides host-to-device memory copy time without any code modification. It overlaps the host-to-device memory copy with host computation or CUDA kernel computation by exploiting Unified Memory and fault mechanisms. HUM provides wrapper functions of CUDA commands and executes host-to-device memory copy commands in an asynchronous manner. We also propose two runtime techniques. One checks if it is correct to make the synchronous host-to-device memory copy command asynchronous. If not, HUM makes the host computation or the kernel computation wait until the memory copy completes. The other subdivides consecutive host-to-device memory copy commands into smaller memory copy requests and schedules the requests from different commands in a round-robin manner. As a result, the kernel execution can be scheduled as early as possible to maximize the overlap. We evaluate HUM using 51 applications from Parboil, Rodinia, and CUDA Code Samples and compare their performance under HUM with that of hand-optimized implementations. The evaluation result shows that executing the applications under HUM is, on average, 1.21 times faster than executing them under original CUDA. The speedup is comparable to the average speedup 1.22 of the hand-optimized implementations for Unified Memory.

We consider a sum rate maximization problem with user scheduling in Gaussian MIMO broadcast channels, which is a combinatorial optimization problem. We transform it into a cardinality problem taking into account that selecting inactive users is equivalent to allocating zero power to unselected users. Then, we relax the cardinality constraint by introducing a penalty function to promote sparse power allocation among users. The proposed iterative waterfilling with user selection algorithm is a generalization of the well-known iterative waterfilling that maximizes the sum rate under a sum power constraint in MIMO broadcast channels. Numerical results show that it achieves a very high sum rate with a moderate complexity only proportional to the number of users.

Spectrum sensing is used to perceive the spectral environment over a wide frequency band. The multiple measurement vector (MMV) model can be applied to the spectrum sensing scenario since it enables jointly sparse signal recovery. In this paper, a novel spectrum sensing algorithm, referred to as multiple subspace matching pursuit (MSMP), is proposed to reduce the miss detection and false alarm events in the spectrum sensing. Numerical simulations demonstrate that the proposed algorithm shows the outstanding recovery performance with the reduction of the incorrect spectrum decisions.

In this study, the authors investigate the delay‐guaranteed resource optimisation in orthogonal frequency division multiple access networks under time‐varying channels and bursty data arrivals. Stochastic optimisation model is employed to minimise the long‐time‐average transmit power consumption (PC) of base station under the constraints of network stability and individual user's delay requirement. They develop a delay‐guaranteed power‐optimal algorithm (DPOA) to obtain the optimal decisions of stochastic optimisation problem. Without prior knowledge of channel statistics and data arrivals, DPOA yields a time‐averaged transmit PC that can arbitrarily approach the theoretical optimum attained by the network with complete knowledge of statistics. Simulations results verify the theoretical analysis on the network performance and show the effectiveness of DPOA.

In this study, the authors investigate the resource allocation for orthogonal frequency‐division multiple access (OFDMA) wireless networks, where the base station is powered by renewable energy and electric grid. To fully exploit the renewable energy, the authors propose an energy‐aware resource allocation (EARA) algorithm to maximise the network utility, which captures the tradeoff between the system throughput and the grid energy consumption. Specifically, the EARA algorithm only has to track the current system states (e.g. channel and queueing conditions) without requiring a relevant priori distribution knowledge, making it applicable for practical OFDMA wireless networks with unpredictable channel dynamics, renewable energy arrivals and stochastic traffics. Moreover, the performance achieved by the EARA algorithm is theoretically characterised. Most importantly, the authors develop an implementation architecture to take the EARA algorithm into practice, and also analyse the low implementation costs (e.g. low computational complexity, trivial signalling overhead etc.). Finally, simulation results verify the theoretical analysis and also demonstrate the advantages of the EARA algorithm.

We present sufficient conditions under which the generalized orthogonal matching pursuit algorithm recovers the true support set in the noiseless and noisy compressive sensing problems. We derive new bounds of inner products and norms of sparse signals using restricted isometry constants. The proposed sufficient conditions for the successful signal recovery are improved sufficient conditions over the existing ones.

Within a macrocell with a large coverage area, multiple small cells are deployed such that each small cell base station (SBS) supports wireless service demands from user equipments (UEs). Each UE can be simultaneously served by multiple SBSs for quality of service (QoS) enhancement. When there exist hotspot areas with a number of UEs, the SBSs near the hotspot areas may experience a higher resource utilisation level than those outside of the hotspot areas, resulting in a shortage of available resources. The authors propose a robust resource‐utilisation‐based coordinated transmission for heterogeneous networks with a locally different level of traffic demands. In the utilisation‐based coordinated transmission, low‐utilisation SBSs with a small number of UEs are selected to serve a newly joining UE because they have more capacity to serve requests with bursty traffic demand. They further formulate the selection of cooperative SBSs as a robust optimisation problem in order to ensure that UEs have sufficiently high signal‐to‐interference‐plus‐noise ratios, even with channel estimation inaccuracy and strong interference from non‐cooperative SBSs. The simulation results indicate that the proposed method guarantees robust and efficient service performance in heterogeneous small cell networks.

We consider two-user MIMO broadcast and interference channels in which confidential messages intended for each receiver are kept secret from the other receiver. The confidential secrecy rates are nonconvex functions of transmit covariance matrices, which makes it intractable to find optimal transmit covariance matrices analytically. In the MIMO broadcast channels, we prove the weighted secrecy sum rate maximization has zero duality gap and KKT conditions are necessary conditions for the optimal solution. We apply the block successive lower-bound maximization technique to nonconvex weighted secrecy sum rate maximization problems. We also derive numerical algorithms for MIMO interference channels and Gaussian wiretap channels with a cooperative jammer. Every limit point achieved by the proposed algorithms is a local optimal solution that satisfies the KKT conditions. Numerical results show that the proposed algorithms outperform existing suboptimal algorithms in terms of secrecy rates.

In this paper, we present an energy-per-bit minimized radio resource allocation scheme in heterogeneous networks equipped with multi-homing capability, simultaneously connecting to different wireless interfaces. Specifically, we formulate an optimization problem related to minimization of energy-per-bit which takes a form of nonlinear fractional programming. Then we derive a parametric optimization problem out of that fractional programming and solve the original problem by using a double-loop iteration method. In each iteration, we derive the optimal resource allocation policy by applying Lagrangian duality and an efficient dual update method. In addition, we present suboptimal resource allocation algorithms using the properties of the optimal resource allocation policy. Numerical results reveal that the optimal allocation algorithm improves energy efficiency significantly over the existing resource allocation algorithms designed for homogeneous networks and its performance is superior to suboptimal algorithms in reducing energy consumption as well as in enhancing network energy efficiency.

In this paper, we present a resource allocation algorithm for energy consumption minimization in multi-homed wireless networks, simultaneously connecting to different wireless interfaces. Specifically, we first formulate a general problem related to minimization of energy consumption while satisfying minimum rate requirements. Then, we solve the problem by relaxing integer value constraints and applying Lagrangian Lagrangian dualityduality. For a given set of dual variables, we derive optimal resource allocation policies by using an efficient dual update method. Numerical results reveal that the optimal allocation algorithm outperforms the existing resource allocation algorithms for homogeneous networks in terms of the energy consumption reduction as well as energy efficiency of the overall network.

In this paper, we propose a new feedback scheme called scheduling probability prediction-based selective feedback (SPP-SF) for channel state feedback in OFDMA downlink system. In the SPP-SF scheme, each user first calculates the scheduling probability of each subchannel (i.e., the probability that the user wins the scheduling competition for the subchannel) and then reports the state of a part of the subchannels whose “scheduling probabilities” are higher than the other subchannels. The conventional selective feedback (SF) scheme which reports a part of the subchannels whose “MCS levels” are higher than the other subchannels may be regarded as a subset of the SPP-SF scheme applicable in statistically identical subchannel environment. Numerical results reveal that the SPP-SF scheme achieves significant performance gain over the conventional SF scheme, especially when the statistical characteristics of different subchannels are non-identical.

We study the radio resource management (RRM) in orthogonal frequency division multiple access (OFDMA) involved heterogeneous networks, to maximize the system sum-rate under the proportional user rate constraint. An analytical model which reflects the network access features is presented. Allowing multi-homing access and resource element sharing, the RRM problem can be formulated as a linear programming (LP) problem, and the optimal solution provides upper-bound performance. In order to analyze the network selection criterion, we consider an approximated RRM problem with average resource allocations. Two different multi-homing formulations are used, and both are proven to have the same solution, where the network selection is related to the users' rate ratios between different networks. Then, we propose a low complexity suboptimal RRM algorithm, which is composed of a basic part without multi-homing access and a supplementary part with multi-homing support. The basic part makes single network selection and resource allocations. The network selection algorithm is designed based on the criterion found in the approximated problem. After network selection, an efficient resource allocation algorithm is utilized in the OFDMA network, which can employ the multi-user time and frequency diversity well. If multi-homing is allowed, the supplementary part selects the multi-homing users and reallocates partial resources to further improve the performance. Our simulation results show that the proposed suboptimal algorithm can achieve considerable gains over the previous schemes with minor performance degradation compared with the optimal solution.

We propose a packet scheduling policy called maximum sum backlog (MSB) scheduling in wireless relay networks. It assigns a high priority to a user queues whose packets experienced longer delay until arriving the current queues by choosing a queue whose sum backlog multiplied by the corresponding link capacity is highest. We derive the stability region of the wireless relay system and prove that the MSB scheduling is a throughput-optimal policy. Numerical results show that the MSB scheduling provides a better delay fairness among users than the maximal differential backlog scheduling does.

In this paper, we consider a precoder selection criterion for the maximum-likelihood (ML) detector in limited feedback spatial multiplexing systems. When the ML detector is employed, the exhaustive search for the optimal precoder selection requires very high computational complexity. As a good compromise, we propose a simple precoder selection scheme for the ML detector. It reduces the computational complexity significantly with negligible performance degradation. For the wideband OFDM systems, we extend the precoder selection by introducing a clustering concept in selecting the optimal precoder. Numerical results show that the proposed scheme achieves both the diversity order of the ML detector and the performance enhancement of the limited feedback spatial multiplexing. Moreover, the proposed precoder selection algorithm based on the clustering approach provides enhanced performance compared with conventional interpolation and clustering algorithms.

In this paper, we investigate the capacity region of a multiuser shared channel with time-varying transmission power, whose typical examples are the forward packet data channel (F-PDCH) of IS-2000 1xEV-DV system and the high speed downlink shared channel (HS-DSCH) in 3GPP high speed downlink packet access (HSDPA) system. The concavity of the throughput yields the property that additional power contributes more to a weaker user having a lower channel gain than to a stronger user. Noting this property, we consider a power-balanced policy which tries to balance the received SNR of each user by serving a weaker user with higher transmission power. Then, we establish the equivalence of the power-balanced policy and the Pareto optimal policy which yields a throughput vector at the boundary of the capacity region. This enables to characterize the capacity region explicitly, and also renders an easy means to maximize the total throughput while meeting each user's requirement.

We consider the uplink of a cellular communication system with K users per cell and infinite base stations equally spaced on a line. We consider a conventional system that does not make use of joint cell-site processing and compare delay-limited systems and proportional fair scheduling (PFS) systems in terms of the system spectral efficiency C (bit/s/Hz) versus Eb/N0. PFS performs generally better than the delay-limited system in the regime of low to moderate SNR, but for high SNR the optimal delay-limited system achieves throughput comparable to the PFS system with finite users. The delay-limited system is interference limited. We characterize this limit and validate a commonly made assumption, that outer-cell interference power is equal to a multiplicative factor times the total cell transmit power. In contrast, the spectral efficiency of PFS system can grow without bound if the number of users tends to infinity, thanks to the multiuser diversity effect.

In this paper, we analyze the performance of spacetime codes and propose a distance spectrum computation method in fast Rayleigh fading channels. We first derive a new FER upper bound using the union bound and the PEP upper bound in the fast fading environment. The derived FER upper bound is very accurate, requires only the distance spectrum of the spacetime code, and takes a closed-form expression. Then we propose a distance spectrum computation method of space-time codes in fast fading channels, which exploits the symmetric property of the error state diagram in space-time trellis coded MPSK modulation to reduce the computation complexity. Numerical results illustrate that the derived FER bound is tight enough to estimate the performance of space-time codes in fast fading channels with sufficient accuracy.

In this paper, we investigate the effect of spatial correlation on throughput performance of downlink multi-antenna transmission schemes exploiting multiuser diversity, in which partial channel information such as signal-to-interference plus noise power ratio (SINR) is available at the transmitter. The asymptotic analysis is performed based on the extreme value theory. From this analysis, we demonstrate that the throughput optimal transmission scheme depends on the degree of the antenna correlation and the operating SNR. Especially, the multiuser spatial multiplexing known as the asymptotically optimal transmission scheme is no longer optimal in highly correlated multiple antenna channels.

When a spread spectrum contention-based feedback channel is employed, it has been shown that most of the multiuser diversity gain can be maintained without any delay constraint. To address the delay in sending packets, we investigate how a spread spectrum contention-based feedback channel affects the delay and throughput performance. Using large deviations techniques, we show that most of the multiuser diversity gain can be maintained while satisfying a delay constraint in which the probability that the maximum delay of any bits in any users' queues is less than a specified value. In addition, we show that the spectral efficiency improves as the number of users increases while maintaining a fixed normalized maximum delay (which is defined as the ratio of the maximum delay and the number of users) as the number of users increases.

This paper examines methods of increasing the downlink channel capacity based on transmitting a signal through a multiple antenna system such as beamforming (BF), transmit diversity (TD) and the combined approach integrating BF and TD. A convenient criterion of efficiency of these techniques is proposed, which reflects the capacity improvement. It is shown that the efficiency of using the multiple transmit antenna system depends on two factors - improvement of the reception performance and mitigation of the intra-system interference. The paper provides the comparative assessment of the methods being researched in terms of their efficiency.

Multiuser diversity gain is an effective technique for improving the performance of wireless networks. This gain can be exploited by scheduling the users with the best current channel conditions. However, this kind of scheduling requires that the base station (or access point) knows some kind of channel quality indicator (CQI) information for every user in the system. When the wireless link lacks channel reciprocity, each user must feed back this CQI information to the base station. The required feedback load makes exploiting multiuser diversity extremely difficult when the number of users becomes large. To alleviate this problem, this paper considers a contention-based CQI feedback where only users whose channel gains are larger than a threshold are allowed to transmit their CQI information through a spread-spectrum based contention channel. Considering the capture effect in this contention channel, it is shown that i) the multiuser diversity gain can be exploited regardless of the number of transmit antennas at the base station and ii) the total system throughput exponentially approaches that of the full feedback scheme as the spreading code length of the contention channel linearly increases. In addition, it is also shown that multiuser diversity can be maintained with the feedback delay of time-variant channels. We also consider the issue of differentiated rate scheduling, in which the base station gives different rates to different subsets of mobiles. In this scenario, mobiles feed back their CQI with some access probability, and we show this technique causes only a negligible throughput loss compared to the case without supporting differentiated rate.

In this paper, we provide an efficient quality-of-service (QoS)-guaranteeing scheme using the cdf-based scheduling (CS) algorithm in wireless fading channels. We first extend the CS algorithm such that it can encompass the practical environment with discrete user transmission rates. The extended CS algorithm can allocate the time fractions to users in an arbitrary manner, and render an exact estimation of user average throughput, through which it can provide differentiated QoS to each user. We also introduce the effective capacity concept to describe the delay-constrained capacity of the CS algorithm, both in time-independent and time-correlated channels. In contrast to other existing scheduling algorithms, the CS algorithm enables calculating the effective capacity analytically, rather than estimating it by measurement on the queuing behavior. Using the effective capacity, we can check the feasibility of the user-specified QoS effectively in wireless time-varying channels.

In this paper, we provide an efficient quality-of-service (QoS)-guarantee scheme using the cumulative-distribution-function-based scheduling (CS) algorithm in wireless fading channels. We first extend the CS algorithm such that it can encompass the practical environment with discrete user transmission rates. The extended CS algorithm can allocate the time fractions to users in arbitrary manner, and render an exact estimation of user average throughputs, through which it can provide differentiated QoS to each user. We also introduce the effective-capacity concept to describe the delay-constrained capacity of the CS algorithm, both in time-independent and time-correlated channels. In contrast to other existing scheduling algorithms, the CS algorithm enables calculating the effective capacity analytically, rather than estimating it by measurement on the queueing behavior. Using the effective capacity, we can check the feasibility of the user-specified QoS effectively in wireless time-varying channels

In this paper, we investigate the capacity region of a multiuser shared channel with time-varying transmission power, whose typical example is the forward packet data channel (F-PDCH) of IS-2000 IxEV-DV system. The concavity of the throughput yields the property that additional power contributes more to a weaker user having a lower channel gain than to a stronger user. Noting this property, we consider a power-balanced policy which tries to balance the received SNR of each user by serving a weaker user with higher transmission power. Then, we establish the equivalence of the power-balanced policy and the Pareto optimal policy which yields a throughput vector at the boundary of the capacity region. This enables to characterize the capacity region explicitly, and also renders an easy means to maximize the total throughput while meeting each user's requirement.

In this paper, we investigate throughput performance of downlink multi-antenna transmission schemes exploiting multiuser diversity. The asymptotic analysis is performed based on the extreme value theory. From this analysis, it is demonstrated that the multiuser spatial multiplexing known as the asymptotically optimal transmission scheme is no longer optimal in highly correlated multiple antenna channels

In this paper, we present a new wireless scheduling algorithm based on the cumulative distribution function (cdf) and its simple modification that limits the maximum starving time. This cdf-based scheduling (CS) algorithm selects the user for transmission based on the cdf of user rates, in such a way that the user whose rate is high enough, but least probable to become higher, is selected first. We prove that the CS algorithm is equivalent to a scheduling algorithm that regards the user rates as independent and identically distributed, and the average throughput of a user is independent of the probability distribution of other users. So, we can evaluate the exact user throughput only if we know the user's own distribution, which is a distinctive feature of this proposed algorithm. In addition, we try a modification on the CS algorithm to limit the maximum starving time, and prove that the modification does not affect the average interservice time. This CS with starving-time limitation (CS-STL) algorithm turns out to limit the maximum starving time at the cost of a negligible throughput loss.

In this paper, we analyze the performance of space-time codes by deriving a new approximation of the frame error probabilities for space-time trellis-coded modulations over quasi-static Rayleigh-fading channels. We take advantage of two techniques, the modified bounding technique and the limiting-before-averaging technique, to tighten the upper bound. In addition, we establish a theorem that enables us to reduce the computation and memory needed to calculate the frame error rate (FER) approximation. The newly derived approximation is very tight, requires only the distance spectrum of the space-time code, and can be computed through single numerical integration. Numerical results exhibit that the new approximation is much closer to the simulation results than other existing bounds are, especially in the case of one receive antenna.

We present a new wireless scheduling algorithm based on the cumulative distribution function (cdf) of user transmission rates and also present a simple modification of it to limit the maximum starving time. This cdf-based scheduling (CS) algorithm selects the user for transmission based on the cdf of user rates in such a way that the user whose rate is high enough but least probable to become higher is selected. It turns out that the cdf-based scheduling algorithm is equivalent to a scheduling algorithm that regards the user rates as independent identically distributed (i.i.d.) and the average throughput of a user is independent of other users' probability distribution. A distinctive feature of this proposed algorithm is that the exact user throughput can be evaluated if the user's own distribution is known. The cdf-based scheduling with starving-time limitation (CS-STL) algorithm turns out not to affect the average inter-service time but to limit the maximum starving time at the cost of a negligible throughput loss.

In this paper, we analyze the performance of space-time codes and propose a distance spectrum computation method in fast Rayleigh fading channels. We first derive a new FER upper bound using the union bound and the PEP upper bound in the fast fading environment. The derived FER upper bound is very accurate, requires only the distance spectrum of the space-time code, and takes a closed-form expression. Then we propose a complexity reduction method of computing the distance spectrum of space-time codes in fast fading channels, which exploits the symmetric property of the error state diagram. Numerical results exhibit that the derived FER bound is tight enough to estimate the performance of space-time codes in fast fading channels with sufficient accuracy.

We first present a new method of determining the upper bounds of maximal eigenvalue of Hermitian positive-definite matrix, which includes the Dembo's (1988) upper bound as a special case. Then we derive the kth-order upper bound /spl Lambda//sub k/ and its first-round iteration /spl Lambda//sub k//sup (1)/ which is much tighter than the Dembo's upper bound /spl Lambda//sub 0/.

We analyze the performance of space-time codes. In particular, we derive an upper bound of the frame error probabilities for space-time trellis coded modulations over quasi-static Rayleigh fading channels. Two techniques, modified bounding and limiting before averaging, are used to derive the new bound. The newly derived upper bound is very accurate, needs only the distance spectrum of the space-time code, and can be computed through single numerical integration. Numerical results show that the new upper bound is much tighter than other existing bounds, especially in the case of one receive antenna.

This paper presents an iterative decoding method in the coded MFSK (multilevel frequency shift keying)/FH-SSMA (frequency hopping-spread spectrum multiple access) system. The kernel of the system, which is a symbol APP (a posteriori probability) calculator, accepts channel values and a priori information from the channel decoder and generates reliability values of the received symbols. The channel decoder also makes reliability outputs using the reliability values from the symbol APP calculator. Iterative decoding is performed by repeating this process. Simulation results reveal that the channel coded system with iterative decoding in the FH-SSMA system can reduce the BER significantly, thereby accommodating more users.