Short-range wireless communication with low-power small-size sensors has been broadly applied in many areas such as in environmental observation, and biomedical and health care monitoring. However, such applications require a wireless sensor operating in "always-on" mode, which increases the power consumption of sensors significantly. Asynchronous communication is an emerging low-power approach for these applications because it provides a larger potential of significant power savings for recording sparse continuous-time signals, a smaller hardware footprint, and a lower circuit complexity compared to Nyquist-based synchronous signal processing. In this dissertation, the classical Nyquist-based synchronous signal sampling is replaced by asynchronous sampling strategies, i.e., sampling via level crossing (LC) sampling and time encoding. Novel forward error correction schemes for sensor communication based on these sampling strategies are proposed, where the dominant errors consist of pulse deletions and insertions, and where encoding is required to take place in an instantaneous fashion. For LC sampling the presented scheme consists of a combination of an outer systematic convolutional code, an embedded inner marker code, and power-efficient frequency-shift keying modulation at the sensor node. Decoding is first obtained via a maximum a-posteriori (MAP) decoder for the inner marker code, which achieves synchronization for the insertion and deletion channel, followed by MAP decoding for the outer convolutional code. By iteratively decoding marker and convolutional codes along with interleaving, a significant reduction in terms of the expected end-to-end distortion between original and reconstructed signals can be obtained compared to non-iterative processing. Besides investigating the rate trade-off between marker and convolutional codes, it is shown that residual redundancy in the asynchronously sampled source signal can be successfully exploited in combination with redundancy only from a marker code. This provides a new low complexity alternative for deletion and insertion error correction compared to using explicit redundancy. For time encoding, only the pulse timing is of relevance at the receiver, and the outer channel code is replaced by a quantizer to represent the relative position of the pulse timing. Numerical simulations show that LC sampling outperforms time encoding in the low to moderate signal-to-noise ratio regime by a large margin. In the second part of this dissertation, a new burst deletion correction scheme tailored to low-latency applications such as high-read/write-speed non-volatile memory is proposed. An exemplary version is given by racetrack memory, where the element of information is stored in a cell, and data reading is performed by many read ports or heads. In order to read the information, multiple cells shift to its closest head in the same direction and at the same speed, which means a block of bits (i.e., a non-binary symbol) are read by multiple heads in parallel during a shift of the cells. If the cells shift more than by one single cell location, it causes consecutive (burst) non-binary symbol deletions. In practical systems, the maximal length of consecutive non-binary deletions is limited. Existing schemes for this scenario leverage non-binary de Bruijn sequences to perfectly locate deletions. In contrast, in this work binary marker patterns in combination with a new soft-decision decoder scheme is proposed. In this scheme, deletions are soft located by assigning a posteriori probabilities for the location of every burst deletion event and are replaced by erasures. Then, the resulting errors are further corrected by an outer channel code. Such a scheme has an advantage over using non-binary de Bruijn sequences that it in general increases the communication rate.
|
![[NJIT logo]](http://archives.njit.edu/vhlib/images/njit_logo.gif){kind=logo}
![[Van Houten Library logo]](images/vhl-top-image.jpg)
