Machine Learning (ML) is a promising tool to design wireless physical layer (PHY) components. It is particularly interesting for millimeter-wave (mm-wave) frequencies and above, due to the more challenging hardware design and channel environment at these frequencies. Rather than building individual ML-components, in this paper, we design an entire ML-enhanced mm-wave receiver for frequency selective channels. Our ML-receiver jointly optimizes the channel estimation, equalization, phase correction and demapper using Convolutional Neural Networks. We also show that for mm-wave systems, the channel varies significantly even over short timescales, requiring frequent channel measurements, and this situation is exacerbated in mobile scenarios. To tackle this, we propose a new ML-channel estimation approach that refreshes the channel state information using the guard intervals (not intended for channel measurements) that are available for every block of symbols in communication packets. To the best of our knowledge, our ML-receiver is the first work to outperform conventional receivers in general scenarios, with simulation results showing up to 7 dB gains. We also provide an experimental validation of the ML-enhanced receiver with a 60 GHz FPGA-based testbed with phased antenna arrays, which shows a throughput increase by a factor of up to 6 over baseline schemes in mobile scenarios.

Mm-wave communications use analog beamforming techniques, which steer the signal energy in a desired direction, to overcome the high path-loss at such frequencies. To determine the direction in which to steer, mm-wave standards such as IEEE 802.11ad specify beam training mechanisms for both access points as well as client stations. However, the overhead of the beam training limits scalability as the density of network deployments increases and mobile devices that require constantre-trainingare supported. We design SPIDER, a low-overhead beam-training mechanism where only access points actively participate in the training and stations perform passive compressive estimation of the angle-of-arrival. To this end, stations carry out phase-coherent measurements by switching through multiple receive beam patterns on a time-scale of tens of nanoseconds when receiving a packet preamble. Since no suitable testbed platforms exist that support such fast antenna reconfiguration, we design a high-performance, full-bandwidth FPGA-based testbed platform for flexible mm-wave experimentation, that we make available as open source. The performance analysis with this testbed shows that our algorithm achieves highly accurate angle estimation used to drive the beam steering decisions and reduces overhead by an order of magnitude compared to IEEE 802.11ad beam training.

Machine learning is a highly promising tool to design the physical layer of wireless communication systems, but its scaling properties for this purpose have not been widely studied. Machine learning algorithms are typically evaluated to learn SISO communications and low modulation orders, whereas current wireless standards use MIMO and high-order modulation schemes to increase capacity. The memory requirements of current Machine learning algorithms for wireless communications increase exponentially with the number of antennas and thus they cannot be used for advanced physical layers and massive MIMO. In this paper, we study the requirements of end-to-end Machine learning models for large-scale MIMO systems, determine the bottlenecks of the architecture, and design different solutions that vastly reduce overhead and allow training higher MIMO and modulation orders. We show that by training the autoencoder in a bit-wise manner, the memory requirements are reduced by several orders of magnitude, which is a critical step for Machine learning-based physical layer design in practical scenarios. Additionally, our design also improves performance over the classical autoencoder for MIMO.

Millimeter-wave systems not only provide high data rates and low latency, but the very large bandwidth also allows for highly accurate environment sensing. Such properties are extremely useful for smart factory scenarios. At the same time, reusing existing communication links for passive object localization is significantly more challenging than radar-based approaches due to the sparsity of the millimeter-wave multi-path environment and the weakness of the reflected paths compared to the line-of-sight path. In this paper, we explore the passive object localization accuracy that can be achieved with IEEE 802.11ad devices. We use commercial Access Points (APs) whereas the station design is based on a full-bandwidth 802.11ad compatible FPGA-based platform with a phased antenna array. The stations exploit the preamble of the beam training packets of the APs to obtain Channel Impulse Response (CIR) measurements for all antenna patterns. With this, we determine distance and angle information for the different multi-path components in the environment to passively localize a mobile object. We evaluate our system with multiple APs and a moving robot with a metallic surface. Our system operates in real-time and achieves 6.5cm mean error accuracy and sub-meter accuracy in 100% of the cases.

Millimeter-Wave (mm-wave) technology is increasingly being considered for mobile devices and use cases such as vehicular communication. This requires suitable experimentation platforms to support systems-oriented research to tackle the multitude of problems and challenges of mm-wave communications in such environments. To this end, we introduce mm-FLEX, a flexible and modular open platform with real-time signal processing capabilities that supports a bandwidth of 2 GHz and is compatible with mm-wave standard requirements. mm-FLEX integrates an FPGA-based baseband processor with full-duplex capabilities together with mm-wave RF front-ends and phased antenna arrays that are fully configurable from the processor in real-time. To demonstrate the capabilities of mm-FLEX, we implement a scalable, ultra-fast beam alignment mechanism for IEEE 802.11ad systems. It is based on compressive estimation of the signal's angle-of-arrival by means of switching through multiple receive beam patterns on a nano-second time-scale while receiving a packet preamble. Our implementation is open source and is made publicly available to the research community.

Machine learning is a highly promising tool to design the physical layer of wireless communication systems, but it usually requires that a channel model is known. As data rates increase and wireless transceivers become more complex, the wireless channel, hardware imperfections, and their interactions become more difficult to model and compensate explicitly. New machine learning schemes for the physical layer do not require an explicit model but implicitly learn the end-to-end link including channel characteristics and non-linearities of the system directly from the training data.