Motivated by problems in machine-type wireless communications, we consider codebooks of complex Grassmannian lines in N = 2 m dimensions. Binary Chirp (BC) codebooks of prior art are expanded to codebooks of Binary S...
Motivated by problems in machine-type wireless communications, we consider codebooks of complex Grassmannian lines in N = 2 m dimensions. Binary Chirp (BC) codebooks of prior art are expanded to codebooks of Binary Subspace Chirps (BSSCs), where there is a binary chirp in a subset of the dimensions, while in the remaining dimensions there is a zero. BSSC codebooks have the same minimum distance as BC codebooks, while the cardinality is asymptotically 2.38 times larger. We discuss how BC codebooks can be understood in terms of a subset of the binary symplectic group Sp(2m, 2) in 2m dimensions; Sp(2m, 2) is isomorphic to a quotient group of the Clifford group acting on the codewords in N dimensions. The Bruhat decomposition of Sp(2m, 2) can be described in terms of binary subspaces in m dimensions, with ranks ranging from r = 0 to r = m. We provide a unique parameterization of the decomposition. The BCs arise directly from the full-rank part of the decomposition, while BSSCs are a group code arising from the action of the full group with generic r. The rank of the binary subspace is directly related to the number of zeros (sparsity) in the BSSC. We develop a reconstruction algorithm that finds the correct codeword with O(N log 2 N) complexity, and present performance results in an additive white Gaussian noise scenario.
Machine learning is one of the key building blocks in 5G and beyond [1, 2, 3] spanning a broad range of applications and use cases. In the context of mission-critical applications [2, 4], machine learning models shoul...
详细信息
Machine learning is one of the key building blocks in 5G and beyond [1, 2, 3] spanning a broad range of applications and use cases. In the context of mission-critical applications [2, 4], machine learning models should be trained with fresh data samples that are generated by and dispersed across edge devices (e.g., phones, cars, access points, etc.). Collecting these raw data incurs significant communication overhead, which may violate data privacy. In this regard, federated learning (FL) [5, 6, 7, 8] is a promising communication-efficient and privacy-preserving solution that periodically exchanges local model parameters, without sharing raw data. However, exchanging model parameters is extremely costly under modern deep neural network (NN) architectures that often have a huge number of model parameters. For instance, MobileBERT is a state-of-the-art NN architecture for on-device natural language processing (NLP) tasks, with 25 million parameters corresponding to 96 MB [9]. Training such a model by exchanging the 96 MB payload per communication round is challenging particularly under limited wireless resources. The aforementioned limitation of FL has motivated to the development of federated distillation (FD) [10] based on exchanging only the local model outputs whose dimensions are commonly much smaller than the model sizes (e.g., 10 labels in the MNIST dataset). To illustrate, as shown in Figure 1.1, consider a 2-label classification example wherein each worker in FD runs local iterations with samples having either blue or yellow ground-truth label. For each training sample, the worker generates its prediction output distribution, termed a local logit that is a softmax output vector of the last NN layer activations (e.g., {blue, yellow} = {0.7, 0.3} for a blue sample). At a regular interval, the generated local logits of the worker are averaged per ground-truth label, and uploaded to a parameter server for aggregating and globally averaging the local average logit
One of the features of Long Term Evolution (LTE) is the deployment of femtocells as the underlain cell of a macrocell without any intervention of frequency planning to offload the traffic. In this paper we used Markov...
详细信息
One of the features of Long Term Evolution (LTE) is the deployment of femtocells as the underlain cell of a macrocell without any intervention of frequency planning to offload the traffic. In this paper we used Markov chain to derive the expression of blocking probability for both macro and femtocell in terms of traffic parameters of the network. We developed an analytical model to find the expression of probability of forced termination (FT) using combination of mobility model and probability tree considering low dense femtocellular network. Two different trees were designed: a newly originating call which starts its session in a femtocell and that of in a macro cell. The link parameters of small scale fading of wireless network under Multiple Input Multiple Output (MIMO) are combined with the proposed traffic model to get the probability of FT of a real-life network. A new state transition chain was also developed including its solution for LTE traffic of variable bandwidth (BW) and a comparison was made with Erlang’s traffic model.
Water is a limited resource that is essential for sustaining life but is often wasted. By monitoring water consumption in real time, users become aware of the amount of water that they utilize. This contribution prese...
详细信息
Downlink beamforming techniques with low beamformer training overhead are proposed for joint processing (JP) coordinated multipoint transmission (CoMP). The objective is to maximize the weighted sum rate within joint ...
详细信息
This paper presents the recent results of the design of a novel acoustic rainfall sensing system that is low-cost, portable, and easily deployable, which makes use of the recorded sound produced by the impact of the r...
详细信息
Using conventional approaches to estimating arterial oxygen saturation (i.e., SpO2) for individuals, it was impossible to measure unless the given sensor of the pulse oximeter was attached to the finger. In this study...
详细信息
—A product code with single parity-check component codes can be described via the tools of a multi-kernel polar code, where the rows of the generator matrix are chosen according to the constraints imposed by the prod...
详细信息
An emerging application domain concerning content-based recommender systems provides a better consideration of the semantics behind textual descriptions. Traditional approaches often miss relevant information due to t...
详细信息
Nowadays, smartphones and tablet computers have become progressively essential parts of our life. However, these devices are limited in their computational resources compared to other processing devices such as person...
详细信息
暂无评论