An Adaptive Neural Spike Detector with

Threshold - Locked Loop

BITS Pilani Utsav jana 2017H1400173P

Pilani Campus Abhishek Kumar Singhania 2017H1230222P

BITS Pilani, Pilani Campus

Contents

Introduction
Need for neural Spike detection
Challenges of Spike detection
Threshold Locked Loop
Threshold adjustment Algorithm
Closed Loop Analysis
Implementation
References
2
BITS Pilani, Pilani Campus
Introduction

We present the design of an adaptive neural

spike detector that dynamically adjusts the
spike detection threshold based on the SNR of
the neural data sets based on self learning
architecture

3
BITS Pilani, Pilani Campus
Why Spike sorting and detection is required?

4
BITS Pilani, Pilani Campus
Need for neural Spike detection

Neural prosthetics and other bio implantable devices use the Brain
machine Interface technology. Electroencephalography (EEG) is a
typical example for this.
The detection of neural spike activity is a technical challenge that is
a prerequisite for studying many types of brain function. Measuring
the activity of individual neurons accurately can be difficult due to
large amounts of background noise and the difficulty in
distinguishing the action potentials of one neuron from those of
others in the local area.To extract the activity of every single firing
unit influencing that electrode, we need a process called spike
sorting which includes AP detection and classification.
The presentation discusses the challenges of measuring neural
activity and the basic issues of signal detection and classification.

5
BITS Pilani, Pilani Campus
Challenges of Spike detection

There are two different ways to acquire and analyze

electrophysiological data: (offline sorting) or detect and sort
spikes immediately (during acquisition) and only store the
sorted spikes (real-time online sorting) .
The most widely used technique to detect Action Potentials is
the amplitude threshold crossing. The standard thresh hold
setting is based on the mean value added to several times
(typically 5) the standard deviation . A lot of disadvantages
exist in this conventional fixed amplitude threshold scheme .
Noise sources such as distant neural activity or electrode noise
or offsets can cause a change in threshold levels

6
BITS Pilani, Pilani Campus
Neural Threshold detection
7
BITS Pilani, Pilani Campus
Adaptive Techniques
The approach should be more adaptive compared to the
existing one
This process demands continuous user inputs to update
threshold value .System becomes slow. A straightforward
adaptive approach is necessary .
The goal is to dynamically set an optimal threshold that will
detect true spikes from clusters rejecting background noise
peaks.
This presentation presents an architecture that is able to
perform an adaptive and autonomous spike detection and
independent of changing threshold values

8
BITS Pilani, Pilani Campus
Threshold Locked Loop
The Threshold Locked Loop is an example of
control system using feedback .The TLL adjusts
the detection threshold level by referencing the
feedback from the spike sorter
The thresh-hold adjustment is to increase or
decrease the threshold value by a unit step based
on feedback signal
After the feedback there is an idle time which
allows the threshold value change to take place
9
BITS Pilani, Pilani Campus
Functioning of Locked Loop

10
BITS Pilani, Pilani Campus
Loop Performance Measurement
Loop performance is measured by the percentage of true spike
occurences in all detected spikes within a time period. The
detected spike result is put into the FIFO register.The threshold
is re-computed once the FIFO register is completely filled up.
Better the true spike detection better is the performance

11
BITS Pilani, Pilani Campus
Threshold adjustment algorithm

The proposed algorithm is a self learning adaptive algorithm .

A Finite State Machine (FSM) oversees the execution of the
algorithm. The FSM continuously keeps track of the
performance changes and adjusts the threshold levels
accordingly. It increments or decrements the threshold value
by a unit step size depending upon previous performance. If
the performance drop is larger than a predefined value, or a
valid neuron cluster in the Neural Cache gets deleted, the FSM
initializes the TLL and triggers the base threshold re-
calculation.
The threshold adjustment stops when the improvement in the
noise spike detection is trivial. This is the optimal threshold
value
12
BITS Pilani, Pilani Campus
Contd..

Fig 3:Threshold adjustment Algorithm

13
BITS Pilani, Pilani Campus
Closed loop Analysis

SIMULATION SETUP
For better analysis of the closed loop performance ,we
developed an ideal spike sorting model.

Fig.4- Simulation environment setup

14
BITS Pilani, Pilani Campus
Contd..

8-bits binary fixed-point representation.

unit step size for threshold adjustment is 1/256.
Noisy spike train is supplied to the spike detector.
Detected spike waveforms are sent to spike sorter.
Spike sorter sorts the spikes immediately by
referencing the ground truths, feeding the results back
to the spike detector.

15
BITS Pilani, Pilani Campus
Simulation Results
Relationship between performance and threshold step

Fig.5- Performance/threshold changes using neural data sets with

SNR=0.5 16
BITS Pilani, Pilani Campus
Contd..

Fig.6- Performance/threshold changes using neural data set with

SNR=2
17
BITS Pilani, Pilani Campus
Contd..

Impact on performance and threshold for sudden changes in SNR.

SNR drop causes a sharp performance degradation, triggering the threshold

recalculation to compensate for the change in noise level.
18
BITS Pilani, Pilani Campus
VlSI Implementation
Hardware Architecture (Adaptive spike detector block diagram)
1) Spike FIFO
2) Finite State Machine
3) Power-Gated Adder Tree
4) History FIFO

Fig.8- 19
BITS Pilani, Pilani Campus
Contd..

Table-Summary of power and area.

Most of the area is occupied by the spike FIFO.

The adder tree is mostly power gated and is powered only when
threshold recomputation is required.

Threshold recomputation is not a frequent occurring process.

20
BITS Pilani, Pilani Campus
Conclusion.

Non-Gaussian noise rejection capability and

requires only minimal hardware overhead.

Enables dynamic evaluation of the optimal

threshold value, which results in better spike
sorting accuracy.

21
BITS Pilani, Pilani Campus
References
1: M. S. Lewicki, A review of methods for spike sorting: the detection and
classification of neural action potentials. Network: Comp. Neural Syst.9(4)53-78,
1998.
2: I. Obeid, P. D. Wolf. Evaluation of spike-detection algorithms for a
brain-machine interface application, in IEEE Transactions on
3 : L. S. Smith, Testing spike detection and sorting algorithms using
synthesized noisy spike trains, Neuro-IT workshop on Interoperability of
Simulators, Edinburgh, United Kingdom, 2006.
4 : :An Adaptive Neural Spike Detector with Threshold-Lock Loop, Chung- Ching
Peng, Pawan Sabharwal, Rizwan Bashirullah

22
BITS Pilani, Pilani Campus
Thank you

23
BITS Pilani, Pilani Campus