Mixture model (MM) is a probabilistic framework allows us to define dataset containing K different modes. When each of the modes is associated with a Gaussian distribution, we refer to it as Gaussian MM or GMM. Given a data point x, a GMM may assume the existence of a random index k ϵ {1, ..., K} identifying which Gaussian the particular data is associated with. In a traditional GMM paradigm, it is straightforward to compute in closed-form, the conditional likelihood p(x|k, θ) as well as the responsibility probability p(k| x, θ) describing the distribution weights for each data. Computing the responsibility allows us to retrieve many important statistics of the overall dataset, including the weights of each of the modes/clusters. Modern large datasets are often containing multiple unlabelled modes, such as paintings dataset may contain several styles; fashion images containing several unlabelled categories. In its raw representation, the Euclidean distances between the data (e.g., images) do not allow them to form mixtures naturally, nor it's feasible to compute responsibility distribution analytically, making GMM unable to apply. In this paper, we utilize the generative adversarial network (GAN) framework to achieve a plausible alternative method to compute these probabilities. The key insight is that we compute them at the data's latent space z instead of x. However, this process of z → x is irreversible under GAN which renders the computation of responsibility p(k|x, θ) infeasible. Our paper proposed a novel method to solve it by using a socalled posterior consistency module (PCM). PCM acts like a GAN, except its generator CPCMdoes not output the data, but instead it outputs a distribution to approximate p(k|x, θ). The entire network is trained in an “end-to-end” fashion. Trough these techniques, it allows us to model the dataset of very complex structure using GMM and subsequently to discover interesting properties of an unsupervised dataset, including its segments, as well as generating new “out-distribution” data by smooth linear interpolation across any combinations of the modes in a completely unsupervised manner.

The Generative Adversarial Network (GAN) has achieved remarkable progress in generating synthetic images from text, especially since the use of the attention mechanism. The current state-of-the-art algorithm applies attentions between individual regular-grid regions of an image and words of a sentence. These approaches are sufficient to generate images that contain a single object in its foreground. However, natural languages often involve complex foreground objects and the background may also constitute a variable portion of the generated image. In this case, the regular-grid region based image attention weights may not necessarily concentrate on the intended foreground region(s), which in turn, results in an unnatural looking image. Additionally, individual words such as “a”, “blue” and “shirt” do not necessarily provide a full visual context unless they are applied together. For this reason, in our paper, we proposed a novel method in which we introduced an additional set of natural attentions between object-grid regions and word phrases. The object-grid region is defined by a set of auxiliary bounding boxes. They serve as superior location indicators to where the alignment and attention should be drawn with the word phrases. We perform experiments on the Microsoft Common Objects in Context (MSCOCO) dataset and prove that our proposed approach is capable of generating more realistic images compared with the current state-of-the-art algorithms.

Stochastic Gradient Descent (SGD) has been widely adopted in training Deep Neural networks of various structures. Instead of using a full dataset, a so-called {\itshape mini-batch} is selected during each gradient descent iteration. This aims to speed up the learning when a large number of training data is present. Without the knowledge of its true underlying distribution, one often samples the data indices uniformly. Recently, researchers applied a diversified mini-batch selection scheme through the use of Determinantal Point Process (DPP), in order to avoid having highly correlated samples in one batch ({{Zhang et al.}} ({2017})). Despite its success, the attempts were restrictive in the sense that they used fixed features to construct the Gram-matrix for DPP; using the raw or fixed higher-layer features limited the amount of potential improvement over the convergence rate. In this paper, we instead proposed to use variable higher-layer features which are updated at each iteration when the parameter changes. To avoid the high computation cost, several contributions have been made to speed up the computation of DPP sampling, including: (1) using hierarchical sampling to break down a single DPP sampling with large Gram-matrix into many DPP samplings of much smaller Gram-matrix and (2) using Markov k-DPP to encourage diversity across iterations. Empirical results show a much more diversified mini batch in each iteration in addition to a much improved convergence compared with the previous approach.

Spectrum cartography is the process of constructing a map showing Radio Frequency signal strength over a finite geographical area. In our previous work we formulated spectrum cartography as a compressive sensing problem and we illustrated how cartography can be used in the context of discovering spectrum holes in space that can be exploited locally in cognitive radio networks. This paper investigates the performance of compressive sensing based approach to cartography in a fading environment where realtime channel estimation is not feasible. To accommodate for lack of channel information we take an iterative approach. We extend the well-known iteratively reweighted ℓ1minimisation approach by exploiting spatial correlation between two points in space. We evaluate the performance in an urban environment where Rayleigh fading is prominent. Our numerical results show a significant improvement in the probability of accurately making a spectrum sensing decision, in comparison to the well-known weighted approach and the traditional compressive sensing based method.

Licence-exempt secondary Long Term Evolution systems have been proposed recently, in attempt to meet the needs of rapidly growing wireless mobile applications. However, where the secondary network is spread over a large geographical area, traditional detect-and-avoid algorithms are less effective in providing interference protection to Primary Users while maximising the secondary throughput. Spectrum cartography is an emerging technique that can be used to discover spectrum holes in space. We propose a downlink power allocation algorithm using Kriging Spatial Interpolation and Compressive Sensing based spectrum cartography in an environment where large scale shadow fading is prominent. We evaluate the performance of our approach by simulating a secondary Urban Microcell network operating in TV White Space. Simulation results show a significant improvement in interference and throughput, in comparison to traditional detect-and-avoid algorithms.

Spectrum cartography is the process of constructing a map showing Radio Frequency signal strength over a finite geographical area. Multiple research groups have recently proposed to use spectrum cartography in the context of discovering spectrum holes in space that can be exploited locally in cognitive radio networks. In our novel approach, we exploit the sparsity of primary users in space to formulate the cartography process as a compressive sensing problem. Further, we present a novel algorithm for solving the cartography problem that builds on the well-known Orthogonal Matching Pursuit algorithm. We evaluate the performance of our approach by simulating a cognitive radio network where primary users are low power wireless microphones. Our simulation results show a significant improvement in reconstruction error, in comparison to two existing compressive sensing based methods.

There has been an emerging interest in non-coherent ultra-wide bandwidth (UWB) communications, particularly for low-data rate applications because of its low-complexity and low-power consumption. However, the presence of narrowband (NB) interference severely degrades the communication performance since the energy of the interfering signals is also collected by the receiver. In this paper, we compare the performance of two non-coherent UWB receiver structures - the autocorrelation receiver (AcR) and the energy detection receiver (EDR) - in terms of the bit error probability (BEP). The AcR is based on the transmitted reference signaling with binary pulse amplitude modulation, while the EDR is based on the binary pulse position modulation. We analyze the BEPs for these two non-coherent systems in a multipath fading channel, both in the absence and presence of NB interference. We consider two cases: a) single NB interferer, where the interfering node is located at a fixed distance from the receiver, and b) multiple NB interferers, where the interfering nodes with the same carrier frequency are scattered according to a spatial Poisson process. Our framework is simple enough to enable a tractable analysis and provide insights that are of value in the design of practical UWB systems subject to interference.

In this paper, we propose a modified two stage Taylor series (TS) method for position estimation in a 3-D environment when either the time-difference-of-arrival (TDOA) or the distance measurements are available. It is aimed to improve the convergence performance of the traditional Taylor series method. Simulation results demonstrate that the modified TS method can improve the position estimation convergence considerably.

This work deals with the development of pre-filtering techniques for low-cost devices using high data rate communications. Many positioning algorithms have been recently revisited in a centralized architecture scenario, where a cheap mobile sensor is surrounded by N base stations. The overview of the system is composed of two blocks: a smoothing filters and a positioning block. In the smoothing filters block, different algorithms such as smoothing filter, recursive least squares and maximum-likelihood are developed to process multiple time-of-arrival measurements before triangulating the position of the mobile sensor. The positioning algorithms are the Taylor series, direct method and spherical interpolation. All in all, it is shown that the recursive processing of multiple measurements at the input of the positioning algorithm improves not only the accuracy of the triangulated position, but also the robustness of the positioning algorithms. We also explain why the very good results given by the maximum- likelihood should only be seen as a lower-bound of the system.

Energy detection receiver is one of the most interesting solution for non-coherent low data rate and low complexity UWB systems. In order to keep the complexity of the receiver structure low the same approach should be used also for ToA (time of arrival) estimation purposes. In this paper, after showing that a generalized maximum likelihood estimator for ToA is a sliding window energy based estimator, an improved version of it that uses partial channel state information is proposed in order to improve the accuracy of the estimation overcoming the hardware limitation related to the minimum integration time of the integration circuit and to the available number of integrators

Lately, impulse-radio (IR) based ultra wideband (UWB) has emerged as a leading candidate for short range data communications with positioning capabilities. This paper examines the performance of time of arrival (TOA) position estimation techniques of an IR UWB non-coherent energy collection receiver. We present the performance of two different algorithms, namely, the threshold-crossing (TC) and the maximum selection (MAX) algorithms in terms of time of arrival (TOA) estimation error in IEEE 802.15.3 channel models. The architectures of TOA MAX and TC algorithms, suitable for implementation into IR-UWB based non-coherent receivers are presented and evaluated in a top-down design methodology using a hardware description language. The goal is to evaluate the UWB IR energy collection receiver with TOA capabilities on a complete simulation platform

This letter focuses on the performance analysis of the decorrelating receiver in multipath Rician faded CDMA channels. M-ary QAM scheme is employed to improve the spectral efficiency. Approximate expressions are first derived for the two performance indexes: the average symbol error rate (SER) and the average bit error rate (BER) when the decorrelating-first receiver perfectly knows the channel information of the user of interest. To achieve desirable closed-form expressions of the SER and the BER, we exploit results in large system analysis and make assumptions of a high signal-to-interference ratio (SIR) and/or a small Rician K-factor. To measure the receiver performance in the practical scenario, we further derive expressions to approximate the average SER and BER of the decorrelating-first scheme with channel uncertainty. Simulation results demonstrate that the analytical results can also be employed to evaluate the performance of the combining-first receiver.

This paper focuses on the performance analysis of the decorrelating multipath combining receiver in multipath Rician faded CDMA channels. M-ary QAM scheme is employed to improve the spectral efficiency. Approximate expressions are first derived for the two performance indexes: the average symbol error rate (SER) and the average bit error rate (BER) when the receiver perfectly knows the channel information of the user of interest. To achieve desirable closed-form expressions of the SER and the BER, we exploit results in large system analysis and make assumptions of a high signal-to-interference ratio (SIR) and/or a small Rician K-factor. To measure the receiver performance in the practical scenario, we further derive expressions to approximate the average SER and BER with channel uncertainty. The pilot-symbol aided linear minimum mean-square channel estimator is considered to obtain an accurate channel estimate and to obtain an expression for the variance of the channel estimation error required for calculating the error rate. All analytical results are compared with simulation results

In this letter, we propose to use repeaters to increase system information capacity in urban areas, where path loss exponents are higher (as much as 3.4 or more). There is at least a 10% increase in coverage area when repeaters are placed in every cell within the network to increase system capacity. The overall system information capacity is doubled at propagation exponent 3.7-3.9.

This letter investigates the performance analysis of a linear minimum mean square error receiver for M-ary quadratic-amplitude modulation in multipath fading channels. Both channel gain variances and instantaneous channel gains of the interferers are considered for receiver implementation. Approximate expressions for symbol and bit error rates are derived only when the receiver knows the channel gain variances of the interferers. In deriving the analytical expressions, we exploit large system analysis and results in single-user multipath combining. The receiver performance and the accuracy of the theoretical results are examined via simulations.

The paper presents the architecture of a low power, low complexity ultra wideband (UWB) transceiver circuit. The circuit is designed for low data rate, low cost applications with built in location and tracking capabilities. The system is based on a non-coherent architecture which enables the receiver to be extremely simple and largely insensitive to the transmitted pulse shape. The circuit presented contains the oscillator, the transmitter, the receiver and the baseband digital signal processing (DSP) block. The oscillator contains the quartz oscillator, a delay locked-loop (DLL) and edge combiner for clock multiplication to generate a 528 MHz timing signal for pulse generation. The transmitter contains a second DLL to fix the delay of the UWB pulse, the UWB pulse generator and antenna. The transmitted UWB pulse is presented. The receiver contains low noise amplifier, variable gain amplifier, squaring circuits, integrators for energy collection, 4 bit analogue-to-digital converters (ADC), digital control logic, integrator and gain selection logic block and detection/bit decision block. The circuits are designed in a 0.35 /spl mu/m Si-Ge BiCMOS process from Austria Microsystems.

In rapidly time varying channels, the channel impulse response estimation might not be successful by the conventional decision directed scheme following the supervised training. The blind sequence detection would be the only solution in these scenarios. In this paper we proposed a new blind sequence detection algorithm, which has a significant robustness against bit ambiguity and whose performance with uncoded sequence is better than that of some existing blind algorithms with coded systems. Bit error performances have been investigated by computer simulation and compared with coded system. A significant gain has been achieved in SNR as much as 6 dB or more.

This paper presents the results of a co-existence study investigating the impact of ultra wideband (UWB) interference on IEEE802.11b and Bluetooth networks. Results are based on the experimental test measurements made at the University of Oulu, Finland using simple high power UWB transmitter prototypes as interference sources. Preliminary results showed that, under the extreme conditions of this experiment, both IEEE802.11b and Bluetooth networks slightly suffer from the existence of several high proximity UWB signals. In our study, several high power UWB transmitters that greatly exceed the FCC radiation regulations have been used, and the measurement settings presents the worst case scenario due to the very short distance between the interferers and the victim system. Effectively our study indicates the use of hundreds of FCC compatible UWB devices at a same space.

This paper presents the results of a co-existence study investigating the impact of ultra wideband (UWB) interference on IEEE802.11b and Bluetooth networks. Results are based on the experimental test measurements made at the University of Oulu, Finland using simple high power UWB transmitter prototypes as interference sources. Preliminary results showed that, under the extreme conditions of this experiment, both IEEE802.11b and Bluetooth networks will slightly suffer from the existence of several high proximity UWB signals. In our study, several high power UWB transmitters that greatly exceed the FCC radiation regulations have been used, and the measurement settings presents the worst case scenario due to the very short distance between the interferers and the victim system. Effectively our study indicates the use of hundreds of FCC compatible UWB devices at a same space.

The convergence behavior of linear parallel interference cancellation is investigated. Especially the so-called ping-pong effect, where the bit-error rate performance is found to oscillate between two different convergence patterns is studied in detail. This effect is shown to be a direct consequence of the extreme eigenvalues of the correlation matrix, allowing for an analytical approach. Intervals for the dominating eigenvalues within which ping-pong effects can occur are specified and illustrated by examples. It is shown that the decision statistic for traditional parallel cancellation will always exhibit oscillating behavior with either short or long codes. Relaxation factors, leading to weighted cancellation, are shown to be effective for alleviating oscillations and ping-pong effects at the expense of convergence rate. Asymptotic analysis for large systems is applied to uncover the convergence behavior for long code systems.

This paper describes the performance of an LMMSE receiver which has been adapted to operate according to the IEEE 802.11 2.4 GHz wireless LAN (WLAN) standard. The LMMSE receiver offers significant theoretical advantages in terms of interference suppression, ability to operate in substantially time-dispersive environments, and operation in non-line of sight conditions. These advantages combine to offer much greater link distances than conventional RAKE + equalizer receiver structures when applied to the short spreading ratio direct sequence spread spectrum (DSSS) WLAN standard. The prototype which has been constructed offers performance very close to the theoretical limits of the receiver whilst operating in 802.11 mode, as well as offering a higher speed proprietary mode.

In this paper we investigate the performance of a CDMA iterative multi-user receiver with non-binary modulation. The receiver incorporates iterative channel estimation and is applied to multipath fading channels. The particular algorithm investigated is an extension of the work presented in Alexander et al., (2000), and Grant et al., (2000). Results are presented based on analysis, simulation and fixed point DSP implementation.

In this paper, the convergence behaviour of linear parallel interference cancellation is investigated. Especially the so-called ping-pong effect where the bit error rate performance is found to oscillate between two different convergence patterns is studied in detail. This effect is shown to be a direct consequence of the extreme eigenvalues of the correlation matrix, allowing for an analytical approach. Intervals for the dominating eigenvalues within which ping-pong effects can occur are specified and illustrated by examples. Relaxation factors, leading to weighted cancellation, are shown to be effective for alleviating oscillations and ping-pong effects at the expense of convergence rate.

The downlink capacity of WCDMA systems is highly important due to the expected asymmetrical nature of the traffic. It is thus of paramount importance that highly efficient downlink receivers are implemented. The conventional RAKE receivers, as are likely to be initially implemented in UMTS networks, suffer significant performance degradation under high traffic loads. A candidate receiver for the downlink of such networks is the LMMSE chip level equaliser. The principal of such receivers is to restore the orthogonality of users from one cell by equalising the received signal from that cell at chip level before despreading. This technique is not only able to suppress the intra-cell multiple access interference (MAI) but may also be used to reduce the inter-cell MAI. This paper addresses the use of the minimum mean-squared error (MMSE) criteria for the construction of a linear MMSE chip equaliser for a WCDMA downlink receiver. Channel coding, as specified in the 3GPP technical specifications, is employed in both single and multi-cell scenarios. Simulation demonstrate such receivers are able to achieve considerable performance gains when compared to the conventional RAKE receiver.

This paper examines the related issues of capacity, coverage and power control in a multi-cell WCDMA network. The network capacity in this case is based on required link quality thresholds. If all users achieve the required link SIR targets for a given percentage of all power control time slots, the network is deemed not to have exceeded its capacity. When this is 50%, the analysis reverts to the conventional mean SIR target performance. For network operators however a substantially higher link quality requirement is likely. In this case, 95% is considered. This means all users must achieve a SIR of greater than or equal to the target value for 95% of the time. The analysis is based on extended models used for IS-95 systems which relies on Gaussian assumptions of interference. This approach gives an insight into the effects of varying traffic types, different allowances for SIR targets and mixed traffic. This paper presents analysis which allows the minimum required power control to be determined based on network load and traffic mix. It also presents analysis which allows the impact of mobile by mobile power control to be examined.

This paper examines the achievable capacity of a direct-sequence spread-spectrum multiple-access system based on an adaptive minimum mean-square-error (MMSE), single-user receiver over a Gaussian channel. The effect of using both trellis and convolutional codes of varying rates and complexity is investigated in relation to the effect on the maximum number of users and the achievable bit-error rate for a given number of users. A system based on a matched filter receiver is presented for the purposes of comparison. It has been found that, over the band-limited channel considered, the use of convolutional coding reduces the total system capacity for the high efficiency MMSE receiver while still providing a coding gain for a small numbers of users. The use of trellis codes however leads to increased system capacity.

This paper reports on a series of measurements carried out at 2 GHz designed to characterise a wide-band wireless local loop (WLL) in a range of environments. The wide-band channel measurements included data taken in both urban and suburban environments, using directional and omni-directional transmitting antennas, with and without movement near the receiver. The transmitters were building or pole-mounted with receivers inside various rooms in a variety of buildings. In all cases, the receiver was inside a building and no line of sight existed.

A new family of complex-valued spreading sequences was proposed for code-division multiple-access (CDMA) systems by Oppermann and Vucetic (1994, 1997), which offer the system designer the choice of a wide range of correlation properties. This letter identifies a subfamily of spreading sequences for use in CDMA systems in which the sequence elements in any set are orthogonal. The subfamily also offers a very wide range of values for both aperiodic auto-correlation (AC) and aperiodic cross-correlation (CC) functions with the added property of having orthogonal elements.

This paper presents a new family of complex valued pseudo-random sequences for use in coded multiple access communication systems. The family offers a very wide range of values for both auto-correlation (AC) and cross-correlation (CC) functions, allowing great flexibility in the selection of characteristics of sequence sets. Based on the measure described in this paper for the mean-square aperiodic AC and CC values of a set of sequences, the correlation properties of sets of these sequences are compared to well-known sequence sets, and it is shown that sets from the new family of sequences have superior qualities. Tables of parameters for various sequence sets are presented to enable the construction and comparison of sets from this new family.

The multipath characteristics of the wireless local loop (WLL) channel were studied over both short and long time periods in urban and sub-urban high-rise and small-house environments. Wideband measurements were performed at a centre frequency of 1.3 GHz with a sliding correlation measurement system using a chip rate of 15 MHz. The receiver was situated at regular cellular base stations, and the transmitter with an omnidirectional antenna was moved to several rooms in different buildings around the base station sites. During the recordings the movement of people in the measurement rooms was restricted. The short-term results show a high degree of similarity in the multipath characteristics of the WLL channel in urban and sub-urban high-rise environments. The maximum measured values of delay spreads for the urban and suburban high-rise environments were 410 ns and 360 ns, respectively. For the sub-urban small-house environment the data indicate lower values of delay spread. The channel was shown to be very slowly fading, being in most situations essentially static over short time periods of the order of several tens of seconds. The variation of the channel characteristics over long time periods was shown to be significant.

This paper presents a model for a wide-band fading channel for terrestrial mobile applications. The model is based on the results of measurements made in a heavily built-up urban environment using a 25 MHz (bandwidth) signal centred at approximately 2.6 GHz. This paper presents measured impulse responses and details the parameter extraction process used to determine the characteristics of the channel. These parameters are used in the channel simulation package and the output of these simulations are compared to the original data.