Page 69 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 3 – Internet of Bio-Nano Things for health applications

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 3

of a single molecule, the authors demonstrated that the signal within a bit‑interval for designing a convexity met‑
controlled drug release mechanism was shown to be bet‑ ric. Moreover, an adaptive threshold was obtained as the
ter than the constant drug release. Moreover, the PDF weighted sum of convexity metrics of the bit‑intervals in
of the dynamic distance between the transmitter and the which the last, the second last, and the third last bit‑1
receiver was used to optimize the release pro ile at the were detected.
transmitter, the time duration of a frame, and the detec‑
tion threshold at the receiver. 3.1.2 Relay‑assisted‑based mobile MC systems

A non‑coherent iterative detection has been proposed in Forster Resonance Energy Transfer (FRET )‑based MMC
10
[144] considering the 1‑D motion of transmitter and re‑ has been proposed in [147] for detecting tumors. In this
ceiver, where the PDF of irst hitting time was obtained work, authors considered two different types of network
in terms of effective diffusion coef icient [117]. In this con iguration: (i) FRET‑based mobile sensor and actuator
work, ISI mitigation was also implemented by subtracting network in which molecular sensing and actuating tasks
the average ISI concentration from the total received sig‑ were carried out at the molecular level, (ii) FRET‑based
nal in the current symbol‑interval. Another non‑coherent mobile ad‑hoc network with source, relay, and destina‑
detection scheme for OOK was proposed in [145], where tion nodes, where the source transmits information to the
the decision metric was calculated as the difference of relay node or destination in a probabilistic manner. De‑
the signal energies received in current and previous bit‑ tection probability and detection time of a single message
intervals. This technique is useful for the scenarios that were analyzed, where message propagation is modeled
experience strong ISI because it makes the energy differ‑ using the Markov‑chain process.
ence positive in the case of bit‑1 and negative in the case
of bit‑0. 3.1.3 MIMO‑based mobile MC systems

In [133], the statistical nature of CIR in the case of MMC
In [148], MIMO‑MC was proposed for both static and dy‑
was investigated, where the mean and variance of the
namic transmitter and receiver scenarios. Before sending
noiseless and noisy received signal were derived. The
a block of data, a training sequence was sent for chan‑
noisy received signal was shown to be Poisson‑Log nor‑
nel estimation. Channel estimation was done using the
mal distributed which could be approximated as Poisson
maximum likelihood CIR and the least square CIR esti‑
distributed in some cases. In this work, the PDF of the dy‑
mators, where the latter was less complex to implement.
namic distance was also found. Moreover, binary hypoth‑
In this work, the training sequence was selected which
esis testing was done at the receiver for three different maximized the Cramer‑Rao bound of the CIR. The au‑
detection thresholds i.e. ixed threshold, half threshold, thors also studied the performance of DFE, MMSE‑DFE,
and optimal threshold according to MAP rule for Poisson and ZF‑DFE equalizers for ISI/ILI mitigation. Moreover,
distributed signals. Simulated results demonstrated that
to avoid cross‑talk, time interleaving was used, where dif‑
the MAP‑based optimal thresholding scheme achieved the
ferent gates at the transmitter released molecules at dif‑
lowest BER and on the other hand, the half threshold
ferent time intervals within a bit duration. For perfor‑
scheme performed the worst.
mance evaluation, the authors showed that MSEs for the
least‑squares CIR estimator and maximum likelihood CIR
In [146], two transmission techniques based on CSK and
estimator were 16 dB and 14 dB, respectively, at a train‑
Manchester coding were presented, where bit‑1 and bit‑0
ing sequence of length 250 bits. Further, with a decrease
were represented as symbol [1 0] and [0 1], respectively.
Moreover, at the receiver, a concentration difference‑ in training sequence length from 250 bits to 16 bits, the
based detection was proposed which used the maximum MSEs for least squares and maximum likelihood CIR esti‑
concentration difference within a bit‑interval for detec‑ mators increased to 25 and 23 dB, respectively.
tion of the transmitted bit using CSK. For Manchester cod‑
ing at the transmitter, the detector employed a decision 3.1.4 Machine‑learning‑based mobile MC
metric based on concentration difference between suc‑ systems
cessive bits and the detection threshold was found using
Various symbol‑by‑symbol and sequence detectors based
the MAP rule.
on neural network 11 for binary and M‑ary amplitude
In [119], OOK modulation has been used at the transmit‑ modulation schemes were proposed in [31]. Specif‑
ter and an adaptive detection scheme based on local con‑ ically, Recurrent Neural Network (RNN), Bidirectional
vexity of the received signal in the case of bit‑1 and con‑ RNN (BRNN), and Sliding BRNN (SBRNN) with LSTM cells
cavity of the received signal in the case of bit‑0 was pro‑ were used, where SBRNN was shown to perform
posed for the receiver. In this work, the average of the well for a coherence time of ≈ 1‑bit duration.
received signal was found where the average was calcu‑
10 FRET is an energy transfer mechanism observed in luorophores.
lated for the two sampling times within which the signal 11 A signi icant advantage of neural network‑based detectors is that they
was convex. For example, 0.1 s and 0.3 s in Fig. 13. This can perform well in time‑varying channels where CSI is dif icult to ob‑
average value was later subtracted from the peak received tain at the receiving node.

© International Telecommunication Union, 2021 57