Page 59 - 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

was chosen if LLR was greater than the upper threshold. gate. Different situations in the system were mapped to
If the LLR was in between the lower and the upper thresh‑ some states of the Markov process. Further, the proba‑
olds, then only one more sample was collected and LLRT bility for each state was calculated and updated. Based
was repeated again. A communication system similar to on these probabilities, the changes in the system were
[85] was proposed in [90]. In this case, multiple receivers monitored. More speci ically, the spreading of changes in
emitted the same type of molecules to convey information the environment (e.g. gradual tumor growth) was studied
to the FC and at the FC, a simple detection based on a con‑ for monitoring. On the other hand, suboptimal detectors
stant threshold was proposed for decoding information based on myopic policy were proposed. For detection, the
bits. This scheme was slightly inferior to the majority rule aim was to minimize the delay in reporting to the FC. Fur‑
considered in [85], which however had low complexity. thermore, binary and non‑binary stopping time scenarios
were investigated.
A communication system based on several nano‑sensors
sending information to the FC was proposed in [91]. This In [95], two maximum likelihood detection schemes (sin‑
system was for detecting the presence or absence of tar‑ gle molecule type and multiple molecule types) at FC
get (malignant tissue) considering the unknown secre‑ based on decode‑and‑forward relaying and an maxi‑
tion rate of biomarkers and unknown location of the ma‑ mum likelihood detection scheme based on amplify‑and‑
lignant tissue. In this work, the LLR test was used at forward relaying was proposed. It was shown that the
the nano‑sensors for detection, whereas the Generalized‑ multiple molecule type scheme outperformed the other
Likelihood Ratio Test (G‑LRT) and Generalized‑Local Op‑ two proposed schemes for the considered cooperative MC
timum Detector (G‑LOD) were proposed at the FC. In con‑ system.
trast to G‑LRT, G‑LOD was less computationally complex
since it maximized the decision variable with respect to 2.1.5 Machine‑learning‑based MC systems
target location only. A MIMO communication system was
proposed in [92] where multiple transmitters sent dif‑ An adaptive threshold‑based detection using an Arti i‑
ferent molecules to the receivers. A signal processing cial Neural Network (ANN) was proposed in [27]. In
scheme to ind the global peak out of various impulse re‑ this work, the Gradient‑Descent algorithm was used for
sponses received at a single receiver was presented. Fur‑ weight optimization. Moreover, elements of the output
ther, an NP test was performed at each of the receivers to vector of the neural network were used as an adaptive
decide in favor of bit‑1 or bit‑0. Finally, the central node threshold for detecting the bits, e.g., if the number of re‑
i.e., FC collected all the individual decisions from each re‑ ceived molecules in jth bit duration was N rx j and o j was
ceiver and made the decision using K out of N rule i.e., if the corresponding output of ANN then N rx j > o j was the
K or more than K receivers decide bit‑1 then FC decides rule for detecting bit‑1 and vice‑versa. In this work, BER
bit‑1. performance better than state‑of‑the‑art detectors was
claimed at a reduced number of transmitted molecules.
On the other hand, three different variations of maximum
likelihood detection were proposed in [93]. In a perfect In [96], curve itting for the received molecules is done,
reporting scenario: i) the FC assesses each observation where the Least Mean Square (LMS) method is used to up‑
by the receiver to calculate the likelihood. This is termed date the polynomial coef icients. Moreover, the weighted
as Full‑maximum likelihood. ii) The FC assigns equal sum of polynomial coef icients is used as the adaptive
weight to all the samples within a bit‑interval at each threshold for detecting the information in the presence
receiver. This is Limited‑maximum likelihood detection. of ISI. Further, Takagi‑Sugeno (TS) fuzzy model has been
In noisy reporting, the FC adds all the observations that employed, which can approximate complex nonlinear sys‑
arrive from different receivers and inds the likelihood. tems with fewer rules and higher modeling accuracy, and
All the receiver’s use single molecule and decode‑and‑ express the local dynamics of each fuzzy rule. The TS
forward strategy to send information to the FC. Hence this model, which is comprised of an optimized structure (the
technique was termed as SD‑maximum likelihood. SD‑ number of rules and inputs), optimized parameters for
maximum likelihood had the worst performance and the membership function. This strategy can minimize the er‑
Full‑maximum likelihood detection performed the best. ror between the fuzzy output and the desired output. This
work also proposed an ANN detector with two hidden lay‑
Abnormality detection and monitoring schemes using ers, which combines the adaptive fuzzy threshold (in the
sensor networks with FC have been presented in [94]. irst hidden layer) and polynomial approximation (in the
Sensors used the OOK modulation scheme to transmit in‑ second hidden layer).
formation to the FC. After collecting observations from
all the sensors, the FC makes a inal decision on the Further, for a CSK‑based modulation scheme, a deep
state of abnormality. Note that this system was based learning‑based detector was proposed in [28] that min‑
on a partially observable Markov decision process (non‑ imized the difference between detected and transmitted
homogeneous Markov model). In this model, the system symbol vectors. The detector had 70 neurons in the irst
state was de ined as the location and time of occurrences hidden layer and 10 neurons in the second hidden layer.
of abnormalities at multiple sites, and how they propa‑ This detector performed well at SNR values of 18 dB and

© International Telecommunication Union, 2021 47