Page 60 - 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

15 dB for ISI and ISI free scenarios respectively. Further‑ In [112], binary and 4‑ary amplitude modulation schemes
more, to reduce the number of hidden layer neurons, a have been presented. Apart from these two schemes, au‑
different architecture of the feed‑forward ANN detector thors also proposed a third modulation scheme where
has been proposed in [29] where the training of a neural symbols 01 and 10 were considered as 1, and 00 and
network was carried out for zero‑bit, one‑bit, and K‑bit 11 were considered as 0 and 2, respectively. For each
memory receivers using the number of received particles scheme, Symbol Error Probability (SEP) expression was
and the corresponding transmitted bit. The number of derived considering maximum likelihood detection at the
hidden layers was 10 with 5 neurons in each layer. More‑ receiver. Authors demonstrated that in positive drift sce‑
over, the Levenberg‑Marquardt (LM) optimization algo‑ nario, SEP for all three schemes decreased with an in‑
rithm has been used for updating the weights. The detec‑ crease in low velocity. Moreover, SEP performance of the
tion was based on whether the probability of observing third modulation scheme was in between the other two
a bit‑1 given the number of received particles and previ‑ modulation schemes. Extending the previous work [99],
ously estimated bits was greater or less than 0.5. the authors in [113] proposed Viterbi sequence detection
to obtain a lower bound on the BER. Further, a family
Further, two different detectors using feed‑forward NN of weighted sum detectors (i.e., MF and equal weight de‑
having one hidden layer of 16 nodes and RNN having one tectors) were also proposed and their BER performances
Long Short‑Term Memory (LSTM) layer of 16 nodes at a were compared with the Viterbi detector. Analytical ex‑
fusion center in a distributed detection system was pro‑ pressions were also derived for different weighted sum
posed in [97]. This system was proposed for a binary hy‑ detectors. Further, ISI mitigation by using the enzymes
pothesis testing to detect the presence or absence of an in the environment was also included to enhance the per‑
abnormality considering OOK‑based transmission. These formance at the receiver.
detectors were trained using the received signal at the
FC from all the sensors and the corresponding hypothe‑ In [100], an OOK modulation scheme was considered for
sis. Adam optimizer and Gradient Descent with momen‑ transmission. However, for detection, the variance of
tum were used for feed‑forward NN and RNN, respec‑ the propagation time of molecules was calculated and
tively. These schemes avoid the requirement of an ana‑ subtracted from the current time to ind the emitting
lytical channel model and CSI estimation particularly for time of the molecules. This detection process was asyn‑
distributed MC systems. It was shown that an RNN de‑ chronous. In this work, the PDF of propagation time
tector performed well than the feed‑forward NN. Table 2 for superior vena cavae at 120 mm/s blood low veloc‑
summarizes the modulation and detection techniques in ity and capillaries at 790 µm were shown. Further, for
static MC without drift in the channel. performance analysis, the error probability expression in
terms of the Chi‑square Cumulative Distribution Function
2.2 Static nano‑machines in low‑induced (CDF) is derived. Finally, the authors showed that the
diffusive channel variance of propagation time PDF was lesser in superior
vena cavae than the capillaries. Therefore, the commu‑
2.2.1 Single transmitter and single receiver‑ nication in superior vena cavae was slightly more reli‑
able than in capillaries. In [101], MoSK modulation was
based MC systems
used for transmission where N molecules of type‑A and
In [98], a MAP rule‑based signal detection scheme with type‑B were released for sending bit‑1 and bit‑0, respec‑
ISI cancellation was presented for binary and M‑ary trans‑ tively. A novel asynchronous detector named Count‑To‑A‑
missions. The ISI cancellation was achieved by subtract‑ Threshold (CTAT) was designed which counted both types
ing the expected ISI molecules from the number of re‑ of molecules and then decided in favor of a particular bit
ceived molecules in the current bit‑interval. In this work, if the number of molecules corresponding to that bit was
positive drift velocity has been assumed and the PDF of above a prede ined threshold. In the presence of ISI, this
irst hitting time of a molecule is considered as IG. The detector resulted in a better performance than the simple
maximum likelihood sequence detection along with two binary detector, which compared the number of type‑A
variations of a weighted sum detector i.e., equal weight received molecules with type‑B received molecules.
detector and MF detector were proposed in [99] for differ‑ The work in [103] proposed a binary timing‑based mod‑
ent types of low present in the environment. It is shown ulation scheme for conveying the information between
that the maximum likelihood sequence detection outper‑ two communicating nodes. Also, three different detection
formed the other two detectors for most of the scenar‑ schemes were proposed at the receiver. The irst detec‑
ios. However, both weighted detectors performed well if tion scheme was based on MAP criteria where a condi‑
a mild low was present opposite to the direction of infor‑ tional PDF of output was maximized for detection; how‑
mation low. Further, the work in [105] considered OOK ever, it was very complex as the multiplication of several
modulation and proposed an MF detector which maxi‑ IG distributions were required. The second technique was
mizes the SINR to calculate the optimal weights to be
multiplied by each sample within a bit‑interval.
However, the selection of the detection threshold was not
specified therein.

48 © International Telecommunication Union, 2021