TinyML: Tools, Applications, Challenges, and Future Research Directions

In recent years, Artificial Intelligence (AI) and Machine learning (ML) have gained significant interest from both, industry and academia. Notably, conventional ML techniques require enormous amounts of power to meet the desired accuracy, which has limited their use mainly to high-capability devices such as network nodes. However, with many advancements in technologies such as the Internet of Things (IoT) and edge computing, it is desirable to incorporate ML techniques into resource-constrained embedded devices for distributed and ubiquitous intelligence. This has motivated the emergence of the TinyML paradigm which is an embedded ML technique that enables ML applications on multiple cheap, resource- and power-constrained devices. However, during this transition towards appropriate implementation of the TinyML technology, multiple challenges such as processing capacity optimization, improved reliability, and maintenance of learning models' accuracy require timely solutions. In this article, various avenues available for TinyML implementation are reviewed. Firstly, a background of TinyML is provided, followed by detailed discussions on various tools supporting TinyML. Then, state-of-art applications of TinyML using advanced technologies are detailed. Lastly, various research challenges and future directions are identified.

Energy-Efficient Wake-Up Signalling for Machine-Type Devices Based on Traffic-Aware Long-Short Term Memory Prediction

Reducing energy consumption is a pressing issue in low-power machine-type communication (MTC) networks. In this regard, the Wake-up Signal (WuS) technology, which aims to minimize the energy consumed by the radio interface of the machine-type devices (MTDs), stands as a promising solution. However, state-of-the-art WuS mechanisms use static operational parameters, so they cannot efficiently adapt to the system dynamics. To overcome this, we design a simple but efficient neural network to predict MTC traffic patterns and configure WuS accordingly. Our proposed forecasting WuS (FWuS) leverages an accurate long-short term memory (LSTM)- based traffic prediction that allows extending the sleep time of MTDs by avoiding frequent page monitoring occasions in idle state. Simulation results show the effectiveness of our approach. The traffic prediction errors are shown to be below 4%, being false alarm and miss-detection probabilities respectively below 8.8% and 1.3%. In terms of energy consumption reduction, FWuS can outperform the best benchmark mechanism in up to 32%. Finally, we certify the ability of FWuS to dynamically adapt to traffic density changes, promoting low-power MTC scalability

Massive Wireless Energy Transfer with Statistical CSI Beamforming

Wireless energy transfer (WET) is a promising solution to enable massive machine-type communications (mMTC) with low-complexity and low-powered wireless devices. Given the energy restrictions of the devices, instant channel state information at the transmitter (CSIT) is not expected to be available in practical WET-enabled mMTC. However, because it is common that the terminals appear spatially clustered, some degree of spatial correlation between their channels to the base station (BS) is expected to occur. The paper considers a massive antenna array at the BS for WET that only has access to i) the first and second order statistics of the Rician channel component of the multiple-input multiple-output (MIMO) channel and also to ii) the line-of-sight MIMO component. The optimal precoding scheme that maximizes the total energy available to the single-antenna devices is derived considering a continuous alphabet for the precoders, permitting any modulated or deterministic waveform. This may lead to some devices in the clusters being assigned a low fraction of the total available power in the cluster, creating a rather uneven situation among them. Consequently, a fairness criterion is introduced, imposing a minimum amount of power allocated to the terminals. A piece-wise linear harvesting circuit is considered at the terminals, with both saturation and a minimum sensitivity, and a constrained version of the precoder is also proposed by solving a non-linear programming problem. A paramount benefit of the constrained precoder is the encompassment of fairness in the power allocation to the different clusters. Moreover, given the polynomial complexity increase of the proposed unconstrained precoder, and the observed linear gain of the system's available sum-power with an increasing number of antennas at the ULA, the use of massive antenna arrays is desirable.