17-03-2014, 05:36 PM
Abstract—The Wavelet Transform in its discrete form has been
widely applied in the field of biomedical signals. Typically, its
calculation is performed off-line and calculation systems usually
suffer from limited autonomy, bulkiness and obtrusiveness. A
noticeable surge in industrial, research and academic interest
into telemedicine and medical embedded systems, has been
happening recently, where miniature, low-cost, autonomous and
ultra-low-power devices play a major role. Such devices are
usually based on Microcontrollers (MCs) or Field Programmable
Gate Arrays (FPGAs), and in addition to other tasks they need to
perform signal processing, very often in real-time. This paper
presents a methodology to perform on-line QRS detector on
MC’s and FPGA’s platforms. After the theoretical considerations
on wavelets and their optimization in term of integer arithmetic,
the computation architectures for both technologies are
described. At the end, the presentation of obtained results during
intensive tests on real signals is given. Similar approach can be
applied to other signals, where the embedded implementation of
wavelets can be of a benefit.
Keywords-wavelet transform; microcontroller; FPGA; QRS;
I. INTRODUCTION
The Fourier Transform (FT) contains only globally
averaged information and inherently has the drawback to
obscure transient or location specific features within the signal.
This limitation can be remedied to some extent by application
of Short Time Fourier transform (STFT), which uses a sliding
time window of fixed length to localize the analysis. Among
other avialable time–frequency methods, the most promising
seems to be Wavelet Transform (WT).
In contrast to FT, which is restricted to the use of a
sinusoid, WT uses a variety of basic functions, known as
wavelets [1]. In its discrete form (DWT), based on an
orthogonal wavelet, it is particularly useful in signal
compression applications. Another advantage of WT is the
ability to reduce noise in signal, using successively the
procedures of decomposition, thresholding, and signal
reconstruction.
DWT analysis has been broadly applied to a range of
biomedical signals, including ECG, EMG, EEG, PPG, clinical
sounds, respiratory patterns, blood pressure trends and DNA
sequences [2]. Most typically, it is calculated off-line by
custom-design software or general mathematical tools, like
MATLAB or similar. The input data can be prerecorded on
special data buses and formats, like MIT-BIH, or taken from
logger devices, like holters or memory cards. These
circumstances limit the flexibility and transferability of
computation systems.
Nowadays the field of telemedicine and medical embedding
systems are among the fastest growing areas, and the systems
here are based on miniature, low-cost, autonomous and ultralow-
power devices, where MCs and FPGAs perform majority
of tasks. In addition to digitalization, data storage and
communication these devices need to perform real-time signal
processing in all three domains: time, frequency and timefrequency.
It is not a trivial task considering the algorithmic
complexities and limited performances of MCs and FPGAs in
terms of arithmetic power, memory resources, consumption
budget, etc. Undobtedly, the whole this field would
enormously benefit in the optimization of calculation
algorithms, including those DWT based, and their efficient and
effective implementation on MCs and FPGAs platforms.
In this paper we show a way to optimize DWT transform
for its calculation by general purpose, low-cost and low-power
MCs and FPGAs. The principle can be implemented on any
MC, and in our case we selected the MSP430 from Texas
Instruments (TI) as a target platform. In the case of FPGA, the
detector is implemented in Altera’s Cyclone EP1C12Q240
chip.
widely applied in the field of biomedical signals. Typically, its
calculation is performed off-line and calculation systems usually
suffer from limited autonomy, bulkiness and obtrusiveness. A
noticeable surge in industrial, research and academic interest
into telemedicine and medical embedded systems, has been
happening recently, where miniature, low-cost, autonomous and
ultra-low-power devices play a major role. Such devices are
usually based on Microcontrollers (MCs) or Field Programmable
Gate Arrays (FPGAs), and in addition to other tasks they need to
perform signal processing, very often in real-time. This paper
presents a methodology to perform on-line QRS detector on
MC’s and FPGA’s platforms. After the theoretical considerations
on wavelets and their optimization in term of integer arithmetic,
the computation architectures for both technologies are
described. At the end, the presentation of obtained results during
intensive tests on real signals is given. Similar approach can be
applied to other signals, where the embedded implementation of
wavelets can be of a benefit.
Keywords-wavelet transform; microcontroller; FPGA; QRS;
I. INTRODUCTION
The Fourier Transform (FT) contains only globally
averaged information and inherently has the drawback to
obscure transient or location specific features within the signal.
This limitation can be remedied to some extent by application
of Short Time Fourier transform (STFT), which uses a sliding
time window of fixed length to localize the analysis. Among
other avialable time–frequency methods, the most promising
seems to be Wavelet Transform (WT).
In contrast to FT, which is restricted to the use of a
sinusoid, WT uses a variety of basic functions, known as
wavelets [1]. In its discrete form (DWT), based on an
orthogonal wavelet, it is particularly useful in signal
compression applications. Another advantage of WT is the
ability to reduce noise in signal, using successively the
procedures of decomposition, thresholding, and signal
reconstruction.
DWT analysis has been broadly applied to a range of
biomedical signals, including ECG, EMG, EEG, PPG, clinical
sounds, respiratory patterns, blood pressure trends and DNA
sequences [2]. Most typically, it is calculated off-line by
custom-design software or general mathematical tools, like
MATLAB or similar. The input data can be prerecorded on
special data buses and formats, like MIT-BIH, or taken from
logger devices, like holters or memory cards. These
circumstances limit the flexibility and transferability of
computation systems.
Nowadays the field of telemedicine and medical embedding
systems are among the fastest growing areas, and the systems
here are based on miniature, low-cost, autonomous and ultralow-
power devices, where MCs and FPGAs perform majority
of tasks. In addition to digitalization, data storage and
communication these devices need to perform real-time signal
processing in all three domains: time, frequency and timefrequency.
It is not a trivial task considering the algorithmic
complexities and limited performances of MCs and FPGAs in
terms of arithmetic power, memory resources, consumption
budget, etc. Undobtedly, the whole this field would
enormously benefit in the optimization of calculation
algorithms, including those DWT based, and their efficient and
effective implementation on MCs and FPGAs platforms.
In this paper we show a way to optimize DWT transform
for its calculation by general purpose, low-cost and low-power
MCs and FPGAs. The principle can be implemented on any
MC, and in our case we selected the MSP430 from Texas
Instruments (TI) as a target platform. In the case of FPGA, the
detector is implemented in Altera’s Cyclone EP1C12Q240
chip.