Page 64 - 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

bility expression at the destination was derived which was In maximum‑count decoding, one receiver antenna index
shown to be a function of detection thresholds employed is found that receives the maximum number of molecules.
at the relay and destination nodes. Therefore, the authors Corresponding to that antenna index, an n‑bit sequence
determined: (i) the threshold at the relay node by using is decoded where n = log (n t) and n t is the number of
2
the likelihood‑ratio test and (ii) the threshold at the des‑ transmit antennas. In maximum likelihood sequence de‑
tination by minimizing the end‑to‑end error probability. tection, the likelihood function corresponding to a par‑
This end‑to‑end error probability minimization problem ticular symbol sequence is maximized. Towards this,
was formulated as a quasi‑convex optimization problem, the maximum likelihood sequence detector generates the
L
which was solved using the bisection method. trellis to obtain the likelihood of n antenna index vec‑
r
tors, where n r and L represent the number of receiver an‑
The work in [129] proposed an estimate‑and‑forward
tennas and channel memory length, respectively. On the
relaying scheme under the low‑induced diffusive chan‑
other hand, the symbol‑by‑symbol maximum likelihood
nel. In this scheme, estimating the number of transmit‑
detection decodes the symbol by maximizing the sum of
ted molecules at the relay node was done by maximizing likelihood functions at each of the receiver antennas.
the joint PDF of all observations. The Newton‑Raphson
2.2.4 Distributed detection‑based MC systems
method was used for this purpose. Moreover, the analyti‑
cal expression of error probability was derived consider‑
ing maximum likelihood and energy detectors at the re‑ A distributed detection scheme was proposed in [131]
ceiver. Simulation results demonstrated that the decode‑ and [102], where many sensors sent molecules to an FC
and‑forward relaying achieves low BER if the relay is lo‑ to decide the presence or absence of a biological agent
cated close to the source. On the other hand, estimate‑ (BA). In this work, the emission by the BA was modeled as
and‑forward relaying works better if the relay is located the Kolmogorov‑Feller diffusion process (colloidal Brow‑
close to the destination. The work in [109] considered nian motion under drift). The weighted log‑likelihood ra‑
an amplify‑and‑forward relaying scheme to increase the tio test was the optimal fusion rule for binary hypothesis
range of communication. Three different detection tech‑ testing. This rule is called the Chair‑Varshney rule under
niques were presented considering OOK‑based transmis‑ both the NP and Bayesian framework. Further, a subopti‑
sion. In particular, the detection based on mean‑squared mal fusion rule was also presented therein. Event detec‑
error, the detection based on MAP, and the decision rule tion in anomalous diffusion with drift was studied in [132]
which minimized the error probability were considered. for an FC‑based distributed nano‑network. Considering
OOK modulated transmission, the detection at FC was car‑
2.2.3 MIMO‑based MC systems ried out using the binary hypothesis testing with LLR. In
this work, time‑slot optimization based on reinforcement
In [130], the MISO system has been considered where
learning was also proposed to increase the throughput of
multiple transmitters were assumed to send information
the network.
in different time slots using OOK modulation. Apart from
ISI, ILI was also considered while analyzing system per‑ 2.2.5 Machine‑learning‑based MC systems
formance. In this work, symbol duration and the number
of transmitted molecules were jointly optimized. For this High‑dimensional metric combining was proposed in [30]
purpose, multi‑objective optimization was transformed to for a non‑coherent detection scheme. In this work, the
a single objective optimization by using the weighted sum irst metric was constructed using the rising edge prop‑
method. This system was claimed to be useful for drug erty of the signal, the second metric was the minimum
release management where the requirement of various in lection considering successive time slots and the third
drugs in different dosages was present. metric was the energy difference in successive time slots.
At the receiver, the sum of these metrics was used for de‑
Various index modulation schemes utilizing antenna sep‑
tection, which was shown to be insensitive to ISI and im‑
aration were proposed in [107]. The irst scheme con‑
perfect synchronization. Parzen‑probabilistic Neural Net‑
sidered Molecular Space Shift Keying (MSSK) modulation,
work (PNN) was trained using the high‑dimensional met‑
where the transmitter uses each of its antennas to trans‑
ric and corresponding symbols to obtain the approximate
mit different bit‑streams. The second scheme consid‑
likelihood PDFs instead of exact likelihood PDFs for de‑
ered Quadrature Molecular Space Shift Keying (QMSSK)
tection. Also, Parzen‑PNN was claimed to be less complex
where two different MSSK modulators were used at the
than the back‑propagation based ANNs and radial basis
transmitter and each modulator sent a different type
function‑based neural networks.
of molecule thus providing channels orthogonal to each
other. The schematic representation of both MSSK and A non‑linear equalizer, which mitigates ISI and counters
QMSSK schemes are given in Fig. 16. Also, a combi‑ the non‑linearity of the channel has been proposed in
nation of MSSK and QMSSK was proposed as the third [110]. In this work, lower‑to‑higher dimensional map‑
modulation scheme. Three different detection schemes ping was carried out using a machine learning algo‑
i.e., maximum‑count decoding, maximum likelihood se‑
quence detection, and symbol‑by‑symbol maximum
likelihood detection were studied.

52 © International Telecommunication Union, 2021