Page 70 - 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

Besides, a comparison with Viterbi sequence detection 3.2 Mobile nano‑machines in low‑induced
was also presented in [31]. diffusive channel
Further, the work in [149] proposed an ANN‑based detec‑ 3.2.1 Single transmitter and single receiver‑
tion scheme using Broyden–Fletcher–Goldfarb–Shanno based mobile MC systems
(BFGS) algorithm where received signal was iltered then
the iltered signal, slope values and the concentration dif‑ A quaternary modulation scheme for reducing ISI was
ference values of the iltered signal were used for training proposed in [121], where type‑A and type‑B molecules
and detection by the NN. Filtering the received signal was were used for transmission. In this scheme, no molecules
done by partitioning the bit interval into P equal subin‑ were sent for symbol [0 0], type‑A molecules were sent for
tervals and averaging the signal in each partition as de‑ [0 1], type‑B molecules were sent for [1 0], and both types
scribed below. molecules were sent for symbol [1 1]. This scheme was
termed as Depleted‑MoSK (D‑MoSK). For decoding pur‑
jN s +high
1 ∑ poses, maximum likelihood detection was used at the re‑
ϕ i = y j [k], (15)
N p ceiver. Individual thresholds for both types of molecules
k=jN s +low
were used at the receiver e.g. if the number of type‑A
where i = 0, 1, 2, · · · , P −1 and N p is the number of sam‑ molecules was above a certain threshold t a and the num‑
ples within a partition. The low and high parameters are
ber of type‑B molecules was less than the threshold t b
de ined as, low = iN p and high = (i + 1)N p − 1 for then the transmitted symbol was decoded as [0 1].
ϕ i . Using (15), the iltered signal in jth bit interval can be
written as ϕ j ϕ j ϕ j = [ϕ 0 , ϕ 1 , · · · , ϕ P −1 ]. Based on this iltered On similar lines, other symbols were decoded. Simulation
signal, the slope vector s j s j s j and concentration difference results showed that the D‑MoSK modulation scheme re‑
[146] vectord j d j d j can be obtained ass j s j s j = [s 0 , s 1 , · · · , s P −2 ], sulted in a better BER performance than MoSK. Further,
where s i = ϕ i+1 − ϕ i and d j d j d j = [d 0 , d 1 , · · · , d P −2 ], where a novel modulation based on permutation of different
d i = ϕ i+1 − ϕ 0 . Thus, the features z j z j z j (input to the detec‑ molecules was proposed in [153] to achieve better perfor‑
tor) for the three different techniques were based on (a) mance than D‑MoSK in the presence of strong ISI. In this
the iltered signal, (b) slope, and (c) concentration differ‑ modulation scheme, different types of molecules were
ence: sent at different instants within a symbol duration. If M
 different types of molecules are chosen then M! permuta‑
 ϕ j ϕ j ϕ j for (a),
 tions exist and in this case a total ⌊log M!⌋ bits per sym‑
2
z j z j z j = [s 0 , s 1 , · · · , s P −2 , E(s j s j s j ), var(s j s j s j )] for (b), bol can be sent. At the receiver, maximum likelihood de‑


[d 0 , d 1 , · · · , d P −2 , E(d j d j d j ), var(d j d j d j )] for (c), tection was performed. This modulation scheme achieved
(16) better BER performance than MoSK and D‑MoSK schemes
where E(·) and var(·) denote the expectation and the vari‑ for a large ratio of number of transmitted molecules to in‑
ance. In this work, the transmitted bits are represented in formation bits per symbol.
T
the following manner. For bit‑0, p j p j p j = [0 1] and for
T
bit‑1, p j p j p j = [1 0] . Note that the values inside p j p j p j repre‑ In [154], a nano‑sensor network for in‑body applications
sent the Probability Mass Function (PMF) of the transmit‑ has been proposed in which various nano‑machines used
P P
ted bit. If L consecutive bits are transmitted then P L L L = repellent molecules to move away from one another. This
[p 1 p 1 p 1 ,p 2 p 2 p 2 , · · · ,p L L L ] represents the sequence of transmitted process is suitable for searching a target inside the body.
p p
z
bits and Z L L L = [z 1 z 1 z 1 ,z 2 z 2 z 2 , · · · ,z L L z L ] be the corresponding fea‑ Further, attractant molecules were used by the nano‑
Z Z
turesthenthe trainingdata setis representedby(P L L L ,Z L L L ). machines to come closer once the target is detected. In
Z Z
P P
12
The output of the NN is the vector ˆ p p p j j j = NN(z j z j z j ;W W W), this work, a non‑diffusion ‑based MC has been assumed
where W W W represents the weights and biases in the NN. Fi‑ and the gradient of adhesive molecules which could bind
nally, the bits are estimated according to the following cri‑ to the inner wall of the blood vessel were used for guid‑
terion ing the mobile nano‑machines. This kind of coordination
can be useful where loss of molecules due to diffusion is
ˆ b j = arg max b j ϵS ˆ p p p , (17)
j j j high. More speci ically, the aim of this work was to eval‑
T
where ˆ p p p = [Pr(b j = 1|z j z j z j ) Pr(b j = 0|z j z j z j )] is the esti‑ uate the number of nano‑machines near to the target and
j j j
mated PMF vector. the number of nano‑machines hitting the target.
Furthermore, a leader follower‑based MMC has been dis‑
Furthermore, a detection scheme based on Generative‑
Adversarial‑Network (GAN) has been proposed in [150]. cussed in [155]. After inding the target location, the
This network learned from the channel transition prob‑ leader nano‑machines released adhesive molecules which
ability. For ine tuning the transition probabilities, the could be sensed by the follower nano‑machines for mov‑
scheme in [150] combined the use of pilot symbols and ing towards a target (e.g., cancer cells) for various pur‑
generating training data in real time to track time‑varying poses such as drug delivery etc. Maximum likelihood esti‑
channels. Most importantly, this scheme avoids re‑ 12 In non‑diffusion‑based MC, molecules disperse quickly in the environ‑
training the network from the beginning. ment, which has dominant low and limited diffusion.

58 © International Telecommunication Union, 2021