Search

In recent years, deep neural networks have shown to achieve state-of-the-art performance on several classification and prediction tasks. However, these networks demand undesirable lengthy training times coupled with high computational resources (memory, I/O, processing time). In this work, we explore semi-random deep neural networks to achieve near real-time training and less computational resource usage. Although many works enhance the underlying hardware for real-time training, this work focuses on algorithmic optimization. It is shown that random projection networks with additional skipped connectivity and randomly weighted layers can boost the overall network performance while enabling for real-time training. Additionally, a tensor-train decomposition technique is leveraged to further reduce the model complexity of these networks. Our investigation accomplishes the following: 1) Tensor-train decomposition decreases the complexity of random projection networks, 2) compression of the fully connected hidden layer leads to a minimum $\sim40\times$ decrease in memory size, and 3) training under random projection networks can be achieved in near-real time.

This paper presents the design and analysis of orthogonal, Doppler-tolerant waveforms for waveform agile radar (e.g. multiple-input multiple-output (MIMO) radar) applications. Previous work has given little consideration to the design of radar waveforms that remain orthogonal when they are received. Our research is focused on: 1) developing sets of waveforms that are orthogonal on both transmit and receive, and 2) ensuring that these waveforms are Doppler tolerant when properly processed. Our proposed solution achieves the above-mentioned goals by incorporating direct sequence spread spectrum (DSSS) coding techniques on linear frequency modulated (LFM) signals. We call this spread spectrum coded LFM (SSCL) signaling. Our transmitted LFM waveforms are rendered orthogonal with a unique spread spectrum (SS) code. At the receiver, the echo signal will be decoded using its spreading code. In this manner, transmitted orthogonal waveforms can be match filtered only with the intended received signals. From analytical expressions of the waveforms we have designed and from simulation results, we found that: 1) cross-ambiguity function (CAF) of two LFM SS coded (orthogonal) waveforms is small for all delays and Dopplers (i.e. transmit and receive signals satisfy near orthogonality constraint); 2) the length of the SS code determines the amount of interference suppression (i.e., complete orthogonal or near orthogonal of the received signal); 3) we can process the same received signal in two different ways; one method can provide LFM signal resolution and the other method can provide ultrahigh resolution; 4) biorthogonal codes can be used to reduce bandwidth when code length is large. Our proposed waveforms inherit multiple attributes (e.g. chirp diversity, code diversity, frequency diversity) of diverse waveforms.