You are on page 1of 11

A low power mixed signal correlator for

power efficient sound signature detection and

template matching
Mehdi Habibi, Mohammad Shakarami and Ali Asghar Khoddami
Department of Electrical Engineering, University of Isfahan, Isfahan, Iran

Purpose – Sensor networks have found wide applications in the monitoring of environmental events such as temperature, earthquakes, fire and
pollution. A major challenge with sensor network hardware is their limited available energy resource, which makes the low power design of these
sensors important. This paper aims to present a low power sensor which can detect sound waveform signatures.
Design/methodology/approach – A novel mixed signal hardware is presented to correlate the received sound signal with a specific sound signal
template. The architecture uses pulse width modulation and a single bit digital delay line to propagate the input signal over time and analog current
multiplier units to perform template matching with low power usage.
Findings – The proposed method is evaluated for a chainsaw signature detection application in forest environments, under different supply voltage
values, input signal quantization levels and also different template sample points. It is observed that an appropriate combination of these parameters
can optimize the power and accuracy of the presented method.
Originality/value – The proposed mixed signal architecture allows voltage and power reduction compared with conventional methods. A network
of these sensors can be used to detect sound signatures in energy limited environments. Such applications can be found in the detection of chainsaw
and gunshot sounds in forests to prevent illegal logging and hunting activities.
Keywords Sensor networks, Low power, Correlator, Mixed signal circuits, Sound signature
Paper type Research paper

1. Introduction 2014) have been presented. Although signal processing inside

the sensor nodes requires energy, but as radio frequency (RF)
With the advances achieved in sensor network systems and
data transmission is power hungry, raw data transmission is
architectures, the sensing of physical parameters has become
not efficient. If implemented correctly, a wireless node with
possible using arrays of distributed sensor nodes. For example,
signal processing unit can eventually reduce total power rather
sensor networks have been used for impact localization in
than increasing it. Power saving is achieved by several
two-dimensional surfaces (Li et al., 2015). Sensor networks
methods. One technique is to compress data and then send
also have diverse applications in wide area sensing tasks. Using
them over the wireless link (Chakraborty and Tharini, 2015).
sensor networks, it is possible to monitor desired
A lower data bandwidth will require lower power during
measurements or events over broad areas, but an important
transmission. Another effective technique for power usage
challenge with this scenario is that sensor wiring is not possible
reduction is to put the sensor node at a sleep state (Goldberg
over wide fields, and thus, not only sensor data should be
et al., 2006). The low power signal processing stage samples
transmitted using wireless links, but each sensor node should
the environment variable, and when a specific event or
be powered using limited energy resources. Thus, the design
threshold is detected, the sensor is activated and data
of low power sensor units is essential. If the energy
transmission is commenced. A block diagram of such sensor
consumption of a sensor node is low enough, the unit can be
node is shown in Figure 1. In this approach, the design of the
even operated using available environmental energy resources
event detector will be challenging as it can never enter the
such as solar, vibration and electromagnetic sources (Bogue,
sleep state and is always in the active state (Oltean et al.,
2009; Garcia et al., 2015). To this day, many different
environmental sensor nodes such as forest fire detection
Among different applications for distributed sensor network
(Aslan et al., 2012, Bogue, 2013), environmental temperature
systems, recently, the performance of these networks has been
and humidity recorders (Garzón and Riveros, 2010) and
investigated in the field of wide area forest chainsaw detection.
atmosphere pollution monitor systems (Rajasegarar et al.,
Chainsaw event detection is an important audio signal
processing task useful in illegal logging activity prevention
(Czúni and Varga, 2014, Papán et al., 2012, Tang et al.,
The current issue and full text archive of this journal is available on 2012). In recent years, different algorithms have been
Emerald Insight at:

Sensor Review Received 5 June 2016

37/3 (2017) 213–222 Revised 3 September 2016
© Emerald Publishing Limited [ISSN 0260-2288] 25 October 2016
[DOI 10.1108/SR-06-2016-0098] Accepted 28 October 2016

Low power mixed signal correlator Sensor Review
Mehdi Habibi, Mohammad Shakarami and Ali Asghar Khoddami Volume 37 · Number 3 · 2017 · 213–222

proposed which are able to detect specific sound wave signals. which involves certain template matching functions on
The sound wave detection procedure is relatively electrocardiogram (ECG) biomedical signals of the body
process-intensive and usually high computational cost blocks (Gradl et al., 2012). In this case, the low power correlator can
such as wavelet transform, Fourier transform or correlation be advantageous because of the longer battery life of the
are required in the process (Averbuch et al., 2009, George device.
et al., 2013). Many of the works have only focused on the The rest of the paper is organized as follows. In Section 2,
accuracy of the given technique. However, less investigation a review of the template matching technique suitable for signal
has been performed on the power consumption of each signature detection will be presented. In Section 3, the details
method when implemented on a hardware platform. Thus, the of the proposed low power correlator are given. Evaluation
proposal of a suitable hardware which can perform the results are presented in Section 4. Concluding remarks are
detection at adequate accuracy while power consumption is eventually addressed in Section 5.
kept at a minimal level has not been completely addressed and
still remains a challenge. The low power consumption of the
proposed hardware is essential for the distributed sensor
2. Template detection
network implementation of the chainsaw sound detection The correlation procedure is a function in which a signal is
system. This requirement is necessary for a practical correlated with a specific template. The more the input signal
architecture, because as stated earlier, wide area sensor resembles the template waveform, the higher the output
network systems need to be operated with limited energy signal. This procedure is used in many different signal
resources. As it has been shown that correlation-based processing algorithms and tasks such as electroencephalogram
methods are among the relatively accurate detection methods and ECG signal analysis (Ibrahimy et al., 2005) and wavelet
(Chacón-Rodríguez et al., 2011), in this paper, a novel transforms (Ieong et al., 2012). The function for the
hardware is proposed to perform this procedure with low correlation procedure can be described with:
voltage and power requirements. Mixed signal circuits in the

兺 g[m]f[n ⫹ m]
submicron process are used to achieve the desired target. The
fully digital implementation of a correlation block requires [g.f][n] ⫽ (1)
complex and area consuming multipliers. In turn, the high
capacitance effect of the multipliers can lead to substantial
where f is the input signal, g is the template signal, n is the
high dynamic power requirements. In the analog
incoming sample index and j is the number of used template
implementation of the correlation procedure, multipliers
require a small area footprint but suffer from static power
As the equation shows, the input signal f needs to be shifted
dissipation. Also, with the analog approach, the
over time. Although at RF signal frequencies, the continuous
implementation of a long delay line is usually possible through
time shift of a signal is possible with delay time elements
the use of opamps, which are also power hungry. In the
(which can shift the input signal over time at small time
proposed method, a mixed signal approach is used to perform
intervals), the continuous shifting of a low frequency signal is
the correlation procedure using low power requirements. The
cumbersome and discrete signal processing methods are more
delay line is based on a single bit digital shift register while the
appropriate for this purpose. Figure 2 shows how this
multipliers use a time domain analog current multiplier.
procedure is commenced. In Figure 2(a), the input signal
Unlike conventional analog methods, with the given approach,
contains a component similar to the template waveform, while
the lowering of supply voltage and use of submicron
in Figure 2(b), the transient component which has appeared in
technology is possible.
the input signal is not similar to the template waveform. The
The presented low power correlator can also be used in
goal is to detect whether a transient component similar to the
other applications in which signal signature detection is to be
template signal exists in the input waveform. The dashed
performed with moderate accuracy but with limited energy
vertical lines identify the sampling points. For the sake of
resources. In the field of sound signature detection, further
clarity, only the polarity of the shifted input signal and
applications can be found in tasks such as detection of gunshot
template is shown at the sampling points. It is expected that as
sound signatures in forest environments (Chacón-Rodríguez
the input signal is shifted, the highest output at that
et al., 2011). The presented low power correlator can also be
corresponding shift is achieved when the polarities of the input
used for the detection of biomedical signals in implantable
signal and the template signal are the same at their
bio-devices. These devices have limited power sources and
corresponding sampling points. In this case, all the products
they should be powered up using a small battery or by wireless
will be a positive value and their summation will produce a
power transmission. Arrhythmia detection is a sample task
high output result. As it can be seen in Figure 2(a), at time
shift t3, the input signal component has matched the template.
Figure 1 A low power sensor node based on wakeup generation
The polarities of the input signal and the template signal are
upon event detection
similar at their respective sample points, and thus, as
expected, the sum of the products will be a high value at the
Energy source/ Environmental energy harvesster
output waveform in time step t3. When the signal is further
shifted, the output signals will be approximately negative

Sensor frontend Event detector and Wireless

values of previous shift steps. Thus, since a match has
and amplifier wakeup generator transmitter occurred, a non-zero average value will be generated at the
output because of the high output signal value at the match

Low power mixed signal correlator Sensor Review
Mehdi Habibi, Mohammad Shakarami and Ali Asghar Khoddami Volume 37 · Number 3 · 2017 · 213–222

Figure 2 Template matching waveforms

Input signal Output Input signal + Output
signal signal
Delay line Delay line
stage stage
t = t0 - t = t0
+ + t + + t
Template t0 t1 t2 t3 t4 t5 t6 Template t0 t1 t2 t3 t4 t5 t6
- - - -
Input signal Output Input signal Output
+ signal signal
Delay line Delay line
stage stage
t = t1 - t = t1
t - t
+ + + +
Template t0 t1 t2 t3 t4 t5 t6 Template t0 t1 t2 t3 t4 t5 t6
- - - -

Input signal Output Input signal + Output

+ +
signal signal
Delay line Delay line
- stage stage
t = t2 t = t2 -
+ + t + + t
Template t0 t1 t2 t3 t4 t5 t6 Template
t0 t1 t2 t3 t4 t5 t6
- - - -

Input signal Output Input signal Output

+ + signal signal
Delay line Delay line
stage stage
t = t3 - - t = t3 -
t t
+ + + +
Template Template
t0 t1 t2 t3 t4 t5 t6 t0 t1 t2 t3 t4 t5 t6
- - - -

Input signal Output Input signal Output

+ +
signal signal
Delay line Delay line
stage stage
t = t4 - t = t4
- t
+ + t + +
Template t0 t1 t2 t3 t4 t5 t6 Template t0 t1 t2 t3 t4 t5 t6
- - - -

Input signal Output Input signal Output

signal signal
Delay line Delay line
- stage stage
t = t5 t = t5
+ + t + + t
Template t0 t1 t2 t3 t4 t5 t6 Template t0 t1 t2 t3 t4 t5 t6
- - - -

Input signal Output Input signal Output

signal signal
Delay line Delay line
t = t6 - stage
t = t6 stage
+ + t + + t

Template t0 t1 t2 t3 t4 t5 t6 Template t0 t1 t2 t3 t4 t5 t6
- - - -

(a) (b)

Notes: (a) Sample case where a match has occurred between the input signal and the template; (b) Sample case where a match has
not occurred between the input signal and the template

point. In Figure 2(b), however, no match exists and from half (Ibrahimy et al., 2005) domains. In the analog method, for the
of the interval on, the output signal values will negative of delay line, unity gain opamps should be used as voltage
previous values, and thus, the output signal average value will buffers. These opamps usually dissipate significant static
be approximately zero. power, and also for power enhancement, supply voltage
As shown in Figure 3(a), the discrete time correlation can reduction is not possible, because for sufficiently high output
be performed using discrete time shifters, multiplication and voltage swing, the opamp requires enough voltage supply to
summation hardware. As it can be seen from equation (1), the operate.
correlation procedure involves the discrete shifting of the In the digital implementation, shift registers are used to shift
input signal over a delay line. At each step, the respective the binary coded data through the delay line. The main
values of the shifted input signal are multiplied by a static limitation of the digital approach is the need for an analog to
template waveform. The sum of the products gives the output digital converter and also the multibit signal data lines that
signal at that specific shift step. dissipate significant dynamic power. The multibit adders and
As depicted in Figure 3(b) and (c), the correlation function multipliers are also gate-extensive and contribute to significant
can be performed in the analog (Liu et al., 2000) and digital dynamic power usage.

Low power mixed signal correlator Sensor Review
Mehdi Habibi, Mohammad Shakarami and Ali Asghar Khoddami Volume 37 · Number 3 · 2017 · 213–222

Figure 3 (a) Template matching hardware architecture; analog form, and thus, it can be readily connected to a
(b) Implementation of template matching hardware using analog conventional analog threshold detection wakeup circuitry.
circuits; (c) Implementation of template matching using digital Figure 4(a) shows the concept used in the presented
circuits hardware. For an effective correlation, both the template and
the input signals should not have DC components and should
Delay Line
Input vary around a zero value. Dealing with signed number values
δ δ δ
Signal makes the low power hardware design more challenging. The
approach used to overcome this issue is to change the
∑× × × × direction of the current source based on the pulse width
modulation (PWM) signal. Initially, the input signal is
Threshold converted to a digitized pulse width modulated waveform. For
this purpose, an opamp and a sawtooth generator is used. As
the modulated waveform should contain information about
Template Samples
the positive and negative signal values, the sawtooth signal
(a) should oscillate between ⫹VDD and ⫺VDD. Assuming the
j stage input signal is an amplified version of the microphone sensor
which oscillates about the zero value, the produced PWM
+ + +
... signal will have equal high and low periods during a
K1 K2
sufficiently large window time frame. To perform signed
multiplication, the PWM signal is used to change the current
f• g
direction of the multiplier. The multiplicand factors can be
- adjusted by changing the amount of current produced in the
direction-selective current source. Eventually, the produced
current is integrated on a capacitor to detect whether the
g[0] g[1] correlation function has produced a DC component.
To propagate a time-domain PWM signal along a delay
+ +
... line, that signal needs to be sampled over time. Equal to the
K1 K2
C C - C

number of samples per period, D-latch elements are required

(b) in the delay line. The number of samples taken from the PWM
signal will determine the number of input signal quantization
j stage levels. For example, if eight samples are acquired from the
n bit n bit n bit n bit
PWM signal, then the input signal is expressed with eight
f[n] D Q D Q D Q levels of quantization (equivalent to a three-bit binary code),
Binary ...
coded and subsequently, eight D-latch elements will be required in
Clk Clk Clk
the delay line. As one period of the PWM signal only
Clock represents one captured sample of the input signal, thus, for a
... j point correlation, i ⫻ j D-latch units will be required, in


g[0] g[1] g[2]

which i is the quantization levels of the PWM signal.
As shown in Figure 4(b), the multiplier units of the
Adder presented architecture are implemented using direction-
f• g selective current sources. The amount of each current source
(c) determines the corresponding template coefficient. As it can
be seen, the metal– oxide–semiconductor field-effect transistor
dimensions of the current sources are adjusted based on the
3. Proposed method absolute template value. The sign of the template value is
The presented architecture used for sound signature detection applied to the sign terminal. Based on Figure 4(b), the
is based on the conventional correlation technique; however, presented block multiplies the hardwired template coefficient
the hardware used for the processing is different from previous by a factor of ⫹1 or ⫺1 based on the PWM signal state. The
solutions in the fact that mixed signal processing is used in the multiplication is achieved by changing the current signal
procedure. A time-domain digitized signal is used in the delay direction.
line, while an analog current mode circuit is used for the Equation (2) shows how the multiplication is achieved in
multiplier and summation circuit. This approach is effective as one cell using the PWM signal and the direction-selective
the shifting of a signal through a delay line is power efficient current source when the template coefficient has a positive
and can be accomplished by simple shift registers in the digital sign value of I0. During the pulse high time, DT, one
domain. Multipliers on the other hand are hardware-intensive direction of the template current source is selected, and
in the digital domain, and the high number of required gates during the low period, (1-DT), the other direction is
will eventually increase the dynamic power dissipation. Thus, selected. DT[n] is in direct proportion with the sampled
in this work, analog domain multipliers are used for the input signal f[n] (with DT ⫽ 0.5 showing zero); thus, the
correlation technique. Although digitized signals are used in cell output current, Icell, will reflect the multiplication of the
part of the processing, the eventual generated output will be in sampled signal with the corresponding template value:

Low power mixed signal correlator Sensor Review
Mehdi Habibi, Mohammad Shakarami and Ali Asghar Khoddami Volume 37 · Number 3 · 2017 · 213–222

Figure 4 Proposed low power template matching architecture

i stage

coded f[n] 1 bit
D Q D Q ... D Q
1 bit
D Q D Q ... D Q
1 bit
D Q D Q ... D Q
1 bit
Clk Clk Clk Clk Clk Clk Clk Clk Clk

+ - Itotal
Direction C
select g[0] g[1] g[2]
f(t) Saw tooth

j stage


PWM Direction


M5 M3
W M4
L W W |g[m]|
L L I0


g[m] sign
Iref icell

W M2
L W |g[m]|


Notes: (a) Overall architecture. (b) Direction-selective current source used in the proposed architecture

Icell ⫽ DT[n] . I0 ⫹ (1 ⫺ DT[n]) . ⫺ I0 sound signature is mainly produced by a 6,000-18,000 rpm

internal combustion engine which when fired up and
⫽ (2DT[n] ⫺ 1) . I0 (2) accelerated, produces instantaneous frequency components in
the 100-300 Hz range (Potočnik and Poje, 2010) and can be
Considering all current source units, the final current detected by the template matching method. For the
produced on the integration capacitor can be expressed with evaluations, 20 different chainsaw sound tracks with 10-s
equation (3). Thus, the voltage produced on the integration duration, which resembles the chainsaw sound when started
capacitor resembles the integrated value of the correlation and accelerated, are used as the positive sample data. For the
between f and g. If the capacitor voltage value passes the negative data samples, 20 different soundtracks with 10-s
adjusted threshold value, an event will be generated: duration is chosen from natural sound sources such as strong
wind and thunder. These 40 sound sequences are used to

兺 (2DT[n ⫹ m] ⫺ 1). I [m]

Itotal ⫽
evaluate the proposed method. In an ideal scenario, the ratio
of correctly detected chainsaw signals to the total number of

⫽ I 兺 f[n ⫹ m] . g[m]
soundtracks should be as high as possible. This factor will be
ref (3) referred to as the true positive detection ratio. Furthermore,
ideally, the ratio of incorrectly detected chainsaw signals to the
total number of soundtracks should be as low as possible. This
4. Evaluation results will be referred to as the false positive ratio, as it is possible
In this section, different aspects of the proposed sensor will be that the approach incorrectly recognizes other negative
evaluated under different operation conditions. The proposed (non-chainsaw) sound samples as the desired target signal.
design is applied to a 0.18 ␮m standard complementary The true positive, false positive and power consumption are
metal-oxide-semiconductor (CMOS) process. Chainsaw investigated in the evaluations.

Low power mixed signal correlator Sensor Review
Mehdi Habibi, Mohammad Shakarami and Ali Asghar Khoddami Volume 37 · Number 3 · 2017 · 213–222

The criterion on the coverage area has been considered Figure 5 Evaluation of the true positive recognition ratio versus
similar to previous work on chainsaw sound signature different values of input signal quantization levels, i, and different
detection. That is, the maximum distance that a reasonable template sample points, j
chainsaw sound can be picked up by the transducer is
considered about 50 m, in which the sound level of a typical
chainsaw is equal to approximately 60 dB. As the distance 100
increases, the signal to noise ratio of the chainsaw sound
will drop further and recognition ratios will be degraded.

Percentage (%)
Furthermore, only the event detection hardware is 60
investigated in this work, as this is the block that cannot go
to sleep. Thus, it is assumed that when the event is 40
detected, enough energy is accumulated for the wireless 64
transmitter to transmit its result to the base station or
0 16 i
propagate the result through the network (if the data are to
be propagated through the network, additional hardware 9
15 4
will be required for the synchronization of the wireless 18
nodes). j
The variations of the false positive and true positive ratios (a)
are studied with different supply voltage values, different
quantization levels and different template sample points.
Because at process corner conditions and temperatures, the 100
operation of the circuit can be compromised specially at
lower supply voltage values, aside from the different sound 80
Percentage (%)
tracks, the corner conditions and temperatures are also 60
considered as the variable conditions of the evaluations.
Eventually, for each false positive or true positive 40
percentage value, a total number of 240 runs is executed 20
which accounts for 40 different sound tracks under three
0 16 i
different FF, SS, TT corner conditions at two different high
(85°C) and low (⫺10°C) temperatures. The reported 6
power usage values are the average values of these runs for 15 4
each investigated case.
In the evaluations, the target is to reduce the power
consumption as much as possible using the design factors (b)
such as supply potential, delay line length and template
sample points, while keeping the recognition ratios close to Notes: (a) Supply voltage = 1.8 V; (b) supply voltage = 1 V
the values reported in previous work. Initially in the
evaluations, a 150 Hz sinusoid component with 20 ms
number of template sample points is increased, not only
duration is used as the template. However, to further
additional registers are required but also the directional
enhance the recognition ratios, other template waveforms
current multipliers should be extended.
have also been investigated in Section 4.4. Comparison
The procedure of error evaluation is repeated for low
results of Section 4.5 show the successful reduction of
voltage values in Figures 5(b) and 6(b). At low voltage
power usage in relation to previous work.
conditions, the gates are more susceptible to process corner
and temperature variations as the overdrive voltage will
4.1 Delay line length experience larger fluctuations, and furthermore, because of
As stated earlier, the number of quantization levels and the the lower overdrive voltages, the operating speed of the
number of template sample points increase the correlation gates will be reduced. As it can be seen from the results, at
accuracy. In the evaluations shown in Figures 5(a) and higher number of delay line D-latch elements, the error
6(a), the true positives and false positives rates are increases compared with the higher voltage case. This is due
evaluated as the number of quantization levels, and the to the fact that with more delay line registers, a higher clock
template sample points are varied. The minimum value of frequency will be required, and thus, the circuit requires a
the quantization levels and template sample points are higher voltage to be able to support the clock rate.
chosen such that the true positive ratio is higher than 50 per Furthermore, with lower supply voltage, the current
cent and, at the same time, the false positive rate is lower multipliers exit the active region earlier and effect the
than 50 per cent. As it can be seen from the results, the correlator operation, thus reducing accuracy.
increasing of both the quantization levels and template
samples enhances the true positive and false positive rates. 4.2 Power consumption
However, the effect of template sample points is higher than The power consumption of the presented architecture is
the enhancement achieved by increasing the signal evaluated under different values of input signal quantization
quantization levels. It should be noted that when the levels and template sample points. In Figure 7(a), a voltage

Low power mixed signal correlator Sensor Review
Mehdi Habibi, Mohammad Shakarami and Ali Asghar Khoddami Volume 37 · Number 3 · 2017 · 213–222

Figure 6 Evaluation of the false positive recognition ratio versus Figure 7 Evaluation of power consumption versus different values
different values of input signal quantization levels, i, and different of input signal quantization levels, i, and different template sample
template sample points, j points

100 2,000

80 1,500

Power (nW)
Percentage (%)

60 1,000

40 64
20 0 16
16 6 8
0 i 9
6 15 4
9 18
12 4 j 21


Power (nW)
Percentage (%)

60 64
40 0 16
64 6 i
20 32 9
12 4
16 18
0 i j 21
6 8
12 (b)
15 4
j 21
Notes: (a) Supply voltage = 1.8 V; (b) supply voltage = 1 V

Notes: (a) Supply voltage = 1.8 V; (b) supply voltage = 1 V main components. The analog current multipliers mainly
contribute to static power dissipation because of the average
supply of 1.8 V is used. As it can be seen from the figure, to current which they draw from the supply voltage. The delay
achieve higher accuracy by increasing the delay line length, the line has a small static power usage, but because of the high
power usage rises. It should be noted that the increase in number of D-latch elements and their associated capacitance
the delay line length has a two-fold effect on the power usage. effect, the dynamic dissipation of the delay line can reach and
First, the number of elements increases, which results in more even exceed the static power usage of the multipliers especially
power usage and second, with longer delay lines, the clock at higher clock frequencies.
frequency should be increased so that the signal passes Figure 8(a) shows the total power dissipation of the
through the delay line in the same time period as before. The proposed circuit as the supply voltage varies. As the
increasing of the number of template sample points has more evaluation shows, the power usage is reduced as the voltage
effect on the power usage compared with the increasing of drops. This is due to fact that static power, dissipated in
the quantization levels. This is due to the fact that as the the current multipliers, is mainly in direct proportion with
template sample points increases, higher number of current the supply voltage and the dynamic power dissipation in the
multipliers will be required. The evaluation is also repeated digital delay line is mainly in proportion to the second
with low voltage values in Figure 7(b). The voltage reduction power of the supply potential.
can significantly reduce power usage, but as it will be shown in Figure 8(b) shows the performance of the approach in
Section 4.3, the voltage reduction will eventually degrade the detection of chainsaw sound waves by evaluating the true
performance of the circuit. positive and false positive ratios while the supply voltage is
reduced. As the evaluation shows, when the voltage is
4.3 Voltage reduction effect reduced, the correlator shows correct operation for supply
In this section, the effect of voltage reduction is investigated. potentials down to approximately 1 V. To reduce power
With the mixed signal correlation presented in this work and and still sustain correct operation, an operation voltage of
the use of digital delay lines to propagate the sound waveform 1.2 V is chosen for the architecture.
over time, the supply potential can be significantly reduced
compared with conventional techniques without significantly 4.4 Template waveform
affecting the detection performance of the architecture. The To enhance the performance of the presented architecture,
power dissipated in the proposed structure is composed of two the effect of different signature templates is also

Low power mixed signal correlator Sensor Review
Mehdi Habibi, Mohammad Shakarami and Ali Asghar Khoddami Volume 37 · Number 3 · 2017 · 213–222

Figure 8 Performance evaluation of the proposed method versus Figure 9 Evaluation of the proposed hardware using different
supply voltage variations templates extracted from the sound waveforms
False posive rao
True posive rao
Power (nW)


Percentage (%)
600 60
400 40
0.5 0.7 0.9 1.1 1.3 1.5 1.7 1.9 0
Voltage (V)

80 True posive rao
Presentage (%)

False posive rao 8

False posive rao
40 True posive rao

Number of templates
20 5

0 4
0.5 0.7 0.9 1.1 1.3 1.5 1.7 1.9 3
Voltage (V)
Notes: (a) Power consumption; (b) true and false positive 0
recognition ratios
Percentage (%)

investigated in this work. Templates that better resemble (b)

the chainsaw sound waveform can produce more accurate
results. For this reason, the template signal is also chosen Notes: (a) True positive and false positive ratios
based on the chainsaw sound signal itself. Based on the generated by a sin wave and 9 most effective templates;
work of Papán et al. (2012), the templates are chosen at (b) statistical performance of 21 different templates
the segments with the highest auto correlation ratio. In
Figure 9, the false positive and true positive recognition
Varga (2014) is because a general microcontroller is used in
ratios are evaluated based on the templates extracted from
the architecture, while other approaches in the table use
the different chainsaw sound signals. Only the nine most
dedicated low power hardware. For a fair comparison, the
effective templates are illustrated in Figure 9(a). As the
power consumption of the hardware given by
results show, a minimum false positive ratio of 7 per cent
Chacón-Rodríguez et al. (2011) is evaluated with the 1.2 V
and maximum true positive ratio of 92 per cent is achieved
supply voltage used in this work.
in this case. The statistical performance of all the 21
different templates is also given in Figure 9(b), which shows
the number of templates producing a specific true/false 5. Conclusions
recognition ratio. In this paper, a low power sound signature detection
hardware was presented. The hardware is to be used in low
4.5 Comparison with previous work power sensor network architectures which are powered with
For the purpose of comparison, the performance factors of environmental energy resources. By detecting chainsaw
the presented sensor are compared with those obtained in sound waves, the sensor network can be used to detect
previous works (Table I). Although better recognition ratio illegal logging activities in forests and natural habitats. The
values can be achieved using more complex processing, for hardware uses a novel mixed signal correlator with PWM
moderate accuracy and low power tasks, lightweight digital delay line and analog current multipliers. The
processing should be used. As it can be seen from the table, presented architecture is enhanced regarding power usage
the accuracy achieved in this work is comparable to that of by optimizing design parameters such as supply voltage,
previous research. While the accuracy factors of the delay line length and template sample points. The number
presented method remain close to that of previous work, the of required quantization levels and template sample points
power usage is relatively reduced. The high power required for correct operation are extracted in the
consumption of the hardware presented by Czúni and evaluations. With the innovative digital delay line approach,

Low power mixed signal correlator Sensor Review
Mehdi Habibi, Mohammad Shakarami and Ali Asghar Khoddami Volume 37 · Number 3 · 2017 · 213–222

Table I Comparison of the proposed low power sound signature processor with previously presented architectures
Detection Power False True Specified
Reference function consumption positive ratio (%) positive ratio (%) Processor application
Dedicated Forest
Current work Template matching 0.7 ␮W 7 92 hardware chainsaw
Chacón-Rodríguez Template Dedicated Forest
et al. (2011) matching 30 ␮Wⴱ 0ⴱⴱⴱ 91 hardware gunshot
Chacón-Rodríguez Wavelet Dedicated Forest
et al. (2011) transform 4 ␮Wⴱ 6 90 hardware gunshot
Czúni and Varga Forest
(2014) Autocorrelation 7.5 mWⴱⴱ 20 90 Microcontroller chainsaw
Notes: ⴱ Power is reported for 1.2 V supply voltage; ⴱⴱ
Reported power is for both processor and RF transmitter; ⴱⴱⴱ
Reported based on only 15 verifi-
cation samples

the supply voltage can be reduced to a large extent, which center monitoring”, Sensor Review, Vol. 35 No. 4,
can help significantly reduce power usage. The effect of pp. 401-408.
different template signals has also been investigated in the Garzón, C.A.L. and Riveros, O.J.R. (2010), “Temperature,
evaluations. In the optimized state and at a supply voltage humidity and luminescence monitoring system using
of 1.2 V, the correlator shows a false positive detection ratio Wireless Sensor Networks (WSN) in flowers growing”,
of 7 per cent, true positive detection ratio of 92 per cent and ANDESCON, 2010 IEEE, Bogota, pp. 1-4.
power consumption of 0.7 ␮W. While recognition ratios are George, J., Cyril, A., Koshy, B.I. and Mary, L. (2013),
close to those given in previous work, lower power usage is “Exploring sound signature for vehicle detection and
achieved. The presented hardware is a suitable choice for classification using ann”, International Journal on Soft
other sound signature detection tasks such as illegal gunfire Computing, Vol. 4 No. 2, p. 29.
detection in forest environments. The low power correlator Goldberg, D.H., Andreou, A.G., Julián, P., Pouliquen, P.O.,
can also be used for the detection of biomedical templates Riddle, L. and Rosasco, R. (2006), “VLSI implementation
in low power implantable biomedical devices. of an energy-aware wake-up detector for an acoustic
surveillance sensor network”, ACM Transactions on Sensor
References Networks (TOSN), Vol. 2 No. 4, pp. 594-611.
Gradl, S., Kugler, P., Lohmüller, C. and Eskofier, B. (2012),
Aslan, Y.E., Korpeoglu, I. and Ulusoy, Ö. (2012), “A
“Real-time ECG monitoring and arrhythmia detection
framework for use of wireless sensor networks in forest fire
detection and monitoring”, Computers, Environment and using Android-based mobile devices”, 2012 Annual
Urban Systems, Vol. 36 No. 6, pp. 614-625. International Conference of the IEEE Engineering in Medicine
Averbuch, A., Zheludev, V.A., Rabin, N. and Schclar, A. and Biology Society, San Diego, pp. 2452-2455.
(2009), “Wavelet-based acoustic detection of moving Ibrahimy, M.I., Reaz, M.B.I., Mohd-Yasin, F., Khoon, T.
vehicles”, Multidimensional Systems and Signal Processing, and Ismail, A. (2005), “Fetal QRS complex detection
Vol. 20 No. 1, pp. 55-80. algorithm for FPGA implementation”, Web Technologies and
Bogue, R. (2009), “Energy harvesting and wireless sensors: a Internet Commerce, International Conference on Computational
review of recent developments”, Sensor Review, Vol. 29 Intelligence for Modelling, Control and Automation, 2005 and
No. 3, pp. 194-199. International Conference on Intelligent Agents, Vienna, Vol. 1,
Bogue, R. (2013), “Sensors for fire detection”, Sensor Review, pp. 846-850.
Vol. 33 No. 2, pp. 99-103. Ieong, C.I., Mak, P.I., Lam, C.P., Dong, C., Vai, M.I., Mak,
Chacón-Rodríguez, A., Julián, P., Castro, L., Alvarado, P. P.U., Pun, S.H., Wan, F. and Martins, R.P. (2012), “A
and Hernandez, N. (2011), “Evaluation of gunshot 0.83-QRS detection processor using quadratic spline
detection algorithms”, IEEE Transactions on Circuits and wavelet transform for wireless ECG acquisition in
Systems I: Regular Papers, Vol. 58 No. 2, pp. 363-373. 0.35-CMOS”, IEEE Transactions on Biomedical Circuits and
Chakraborty, P. and Tharini, C. (2015), “Analysis of suitable Systems, Vol. 6 No. 6, pp. 586-595.
modulation scheme for compressive sensing algorithm in Li, P., Wang, Y., Hu, J. and Zhou, J. (2015), “Sensors
wireless sensor network”, Sensor Review, Vol. 35 No. 2, distribution optimization for impact localization using
pp. 168-173. NSGA-II”, Sensor Review, Vol. 35 No. 4, pp. 409-418.
Czúni, L. and Varga, P.Z. (2014), “Lightweight acoustic Liu, B.D., Chen, C.Y. and Tsao, J.Y. (2000), “A modular
detection of logging in wireless sensor networks”, The current-mode classifier circuit for template matching
International Conference on Digital Information, Networking, application”, IEEE Transactions on Circuits and Systems II:
and Wireless Communications (DINWC), Hong Kong, Analog and Digital Signal Processing, Vol. 47 No. 2,
pp. 120-125. pp. 145-151.
Garcia, R., Combette, P., Poulin, Y., Foucaran, A., Podlecki, Oltean, G., Grama, L., Ivanciu, L. and Rusu, C. (2015),
J., Hassen, S.B., Grilli, M.A., Hess, O. and Briant, F. “Alarming events detection based on audio signals
(2015), “Piezoelectric energy harvesting: application to data recognition”, 2015 International Conference on Speech

Low power mixed signal correlator Sensor Review
Mehdi Habibi, Mohammad Shakarami and Ali Asghar Khoddami Volume 37 · Number 3 · 2017 · 213–222

Technology and Human-Computer Dialogue (SpeD), of atmospheric pollutants using a system of low-cost
Bucharest, pp. 1-8. sensors”, IEEE Transactions on Geoscience and Remote
Papán, J., Jurecka, M. and Púchyová, J. (2012), “WSN for Sensing, Vol. 52 No. 7, pp. 3823-3832.
forest monitoring to prevent illegal logging”, 2012 Federated Tang, Y., Han, P., Wang, Z., Hu, L., Gao, Y. and Li, H.
Conference on Computer Science and Information Systems (2012), “Based on intelligent voice recognition of forest
(FedCSIS), Wroclaw, pp. 809-812. illegal felling of detecting methods”, 2012 IEEE 2nd
Potočnik, I. and Poje, A. (2010), “Noise pollution in forest International Conference on Cloud Computing and Intelligent
environment due to forest operations”, Croatian Journal of Systems (CCIS), Hangzhou, Vol. 3, pp. 1153-1156.
Forest Engineering, Vol. 31 No. 2, pp. 137-148.
Rajasegarar, S., Havens, T.C., Karunasekera, S., Leckie, C.,
Bezdek, J.C., Jamriska, M., Gunatilaka, A., Skvortsov, A.
Corresponding author
and Palaniswami, M. (2014), “High-resolution monitoring Mehdi Habibi can be contacted at:

For instructions on how to order reprints of this article, please visit our website:
Or contact us for further details:

Reproduced with permission of copyright owner.
Further reproduction prohibited without permission.