Design of Finite Impulse Response Filters for Sampling Rate Conversion

Kantharaj S P
Dept. of E&CE, Bapuji Institute of Engineering and Technology, Davangere, Karnataka, India
Email: kantharajsp@gmail.com
Sunitha G S
Dept. of E&CE, Bapuji Institute of Engineering and Technology, Davangere, Karnataka, India
Email: drsunitha_r_c@yahoo.com
Shilpa K C
Dept. of CS&E, Bapuji Institute of Engineering and Technology, Davangere, Karnataka, India
Email: shilpastjit21@gmail.com
Srikanth M B
Dept. of E&CE, Jain Institute of Technology, Davangere, Karnataka, India
Email: mbshrikant2020@gmail.com
Abstract
As the modern DSP applications are concerned, the increasing requirements of advanced digital
systems are to process data at greater than one sampling rate has resulted in the development of a new sub-
area in the signal processing known as multirate signal processing. Multirate Signal Processing uses different
sampling rates within a system to achieve computational and system efficiencies when compared to system
which operates at single sampling rate. Digital filters in Multirate signal processing are used to suppress
aliasing and to remove imaging effect in decimator and interpolator respectively. Speed and complexity of
multirate filters are dependent on the type, order and design methods of filters. Here, we present the FIR filter
design techniques applicable for sampling rate conversion. The design technique decides the order of the
filter required, which in turn determines the performance parameters of the Multirate systems. In this paper,
single stage decimation and interpolation filter are designed using window techniques and optimal method.
The performance metrics like number of multiplications and storage elements are determined and compared.
For sampling rate converters, and it is proved as required filter order, number of multiplications and storage
elements are less in cases of optimal design methods.

Keywords: Sampling Rate Conversion, Decimation, Interpolation, FIR Filter, Optimal Method.

1. INTRODUCTION

The recent trends such as Software Defined Radio, Internet of Things, Mobile communication have setup a
demand of changing the sampling rate of operating signal, either increasing it or decreasing it by integer or
non-integer factor. The process of converting a given signal sampling rate to a different sampling rate is
called as sampling rate conversion [4, 7]. The systems which operate at different sampling rates throughout
the processing of signals are called as multirate signal processing systems. The Multirate systems perform a
processing task with improved performance metrics and offering that performance at a significantly lower
cost and complexity. Primary operations of sampling rate conversion are Decimation and Interpolation by an
integer factor. Decimation is used to reduce the sampling rate. Whereas interpolation is used to increases the
sampling rate [1, 4, 5].

647
 Input Output
Digital Antialiasing Filter Down Sampler

𝐹𝑠
𝐹𝑠 𝐹𝑠
𝑀

Figure 1. Block Diagram of Decimator

Upsampled
Input signal Output
Up Sampler Digital Antialiasing Filter

𝐿𝐹𝑠
𝐿𝐹𝑠
𝐹𝑠

Figure 2. Block Diagram of Interpolator

Decimation by an integer factor M is the process of retaining the every M th sample of input samples and
hence results in sampling rate of Fs/M. Decimator consists of an anti-aliasing filter fallowed by an M-fold
down sampler as shown in Figure 1. Interpolation by a factor L is the process of inserting (L-1) zero-valued
samples between each input samples hence resulting in the sampling rate of L*fs. Interpolator consists of an
L-fold expander, followed by an anti-image filter as shown in Figure 2 [1, 3].

Filtering is the basic signal operation of sampling rate conversion as it executes direct manipulations of digital
signals, extracts useful frequency components and removes undesirable components as a noise part. In
sampling rate converters, filters are used to suppress aliasing and to remove imaging effect in decimator and
interpolator respectively. Performance of the sampling rate converters depend mainly on these filter
characteristics hence the major challenge is to design the filters to process and transmit the data within desired
frequency band. Contribution of digital filters are significant over the analog filters, hence they preferred for
numerous applications [1, 5].

FIR filters or IIR filters, either of these may be used to generate the overall system response in sampling rate
conversion. Compared to IIR filters, linear phase FIR filters preserve the waveform of the signal components
of interest at the expense of a higher overall complexity. Design techniques for filters are suitably selected
by considering the ideal frequency response which is not practically achievable. Hence, the selection of
approximate design method by the specifications is the major step in the design of filter. The recent literature
reveals that a generalized polyphase FIR filter structure can be used for sampling rate conversion [7].

2. FIR FILTER DESIGN BY WINDOW TECHNIQUES & EQUIRIPPLE METHOD

A FIR filter structure can be used to implement any sort of frequency response for sampling rate
conversions. The non-recursive characteristics of FIR filters yield implementation schemes to improve the
efficiency of the multirate systems significantly. The advantages of FIR filters which are of great
consideration for sampling rate conversion are:
a. Simple and robust filters
b. Easy to obtain linear phase.
c. Implementation of filter is in real time and efficient.
d. More stable and Non-recursive [1, 5].
The important point of consideration of sampling rate converters is to choose the FIR design techniques by
considering the ideal frequency domain and time domain criteria. The design of filter is the process of
determining the filter co efficient which closely approximates the given specifications. In this section we
will discuss the design methods to calculate the FIR filter coefficients applicable to the design of sampling
frequency changing systems.

648
2.1 Window Method
It’s a former and reliable approach to design the FIR filters for the Interpolator and Decimator by
truncating the ideal low pass filter with a certain bandwidth is generated, and then we use a chosen window
to get certain stop band attenuation. Number of windowing techniques are proposed to control the effect of
Gibb’s phenomenon. Selection of the window to be used and the corresponding order is determined with
reference to the Table 1.
Table 1: Windowing Parameters
Window Technique Main lobe / Order Peak Error Side lobe
Rectangular 4π/N -21 dB -13 dB
Hamming 8π/N -44 dB -32 dB
Hanning 8π/N -53 dB -43 dB
Blackman 12π/N -74 dB -58 dB

Two most widely used window designs are the ‘generalized’ Hamming window and Kaiser’s window.
Window techniques truncate the response to be finite with corresponding window functions. The
‘generalized’ Hamming window has function as
2𝜋𝑘 𝑁−1 𝑁−1
𝑤𝐻 (𝑘) = 𝛾 + (1 − 𝛾) cos ( ), − ≤𝑘≤
𝑁 2 2
Where γ is in the range of 0 ≤ γ ≤ 1. If γ = 0.54, then its Hamming window,
If γ = 0.5 then its Hanning window.

Conventionally the order is assumed to be odd so as to produce symmetric coefficients. By adjusting the
order of the filter it is possible to achieve ideal frequency response. If the order of the filter is increased then
width of the main lobe becomes narrower and transition band will be reduced. A simple approximations to
these problems have been developed by Kaiser Window in terms of the Bessel function 𝐼𝑜 (𝑥). In Kaiser
Window technique, length of filter N can be determined with respect to the stop band attenuation and is
adjusted to meet a specified roll-off rate in the transition band. The Kaiser window has the finite function in
the form
(2𝑘)2
𝐼0 (𝛽√1 − ( ))
(𝑁 − 1)2
𝑊𝐾 (𝑘) =
𝐼0 (𝑥)
Where, 𝛽 is the constant that specifies frequency trade off, with A being the attenuation in db, order of the
filter is by Kaiser Window is approximated by
𝐴 − 7.95
𝑁=
14.6 ∆𝑓
2.2 Optimal ( Equiripple ) Method

FIR filters may also be designed by means of more sophisticated design techniques. Optimal filter
design which is derived from chebyshev approximation method designs the FIR filters at the rate of less peak
error. Antialiasing and anti-imaging filters for the Multirate signal processing are designed with respect to
the tolerance scheme of the ideal low pass filter. For sampling rate converter, pass band frequency and stop
band edge frequency are considered according to the specifications to determine the order of the filter. The
important part of the equiripple method is to first estimate the order and then to find the filter co efficient by
the chebyshev approximation. With the given tolerances, Order of the filter for optimal design may be
approximated by

20 log(√𝑅𝑝 𝑅𝑠 − 13)
𝑁=−
14.6 ∆𝑓

Where ∆𝑓 is frequency deviation, Rp and Rs are the ripples in the bands. The Parks-McLellan method solves
for filter coefficients.
As examples of a Sampling rate converters, filters for the single stage decimator and interpolator are designed
for the following specifications and responses are as shown.

649
 Table 2: Sampling Rate Converter’s Specifications
Parameters Decimator Specifications Interpolator Specifications
Sampling frequency 60 kHz 9 kHz
Input frequency 60 kHz 600 Hz
Ripples Rp=0.02, Rs=0.01 Rp=0.002, Rs=0.004
Band edge frequencies fp=1250 Hz, fs=1500 Hz fp=200 Hz, fs=300 Hz
Sampling rate factors 20 15

Linear-phase FIR filters are used in most sampling rate converters as these do not have feedback loops and
are nearly easy to implement. The design constraints and steps for decimator and interpolator are as below.
Step 1: Obtain the sampling rate converters specifications.
Step 2: Select the proper class of filters for single stage conversion.
Step 3: Estimate the filter order from the different methods.
Step 4: Obtain the filter co efficient h(n) and analyse the filter response.
Step 5: Increase filter order if frequency response doesn’t meet ideal response & repeat step 4.
Step 5: Choose the best method of filter design which requires lesser order with better frequency
response.
Step 6: Design the decimator and interpolator as shown in the figures 1 and 2 using the designed FIR
filter.

3. RESULTS AND DISCUSSIONS

In order to design the FIR filters for above sampling rate converters windowing techniques and equiripple
method are used. For windowing techniques, the length of the anti-aliasing filter is estimated with reference
to the Table 1. As another method of designing techniques, optimal equiripple design is used with length of
the FIR filter required is estimated by the order approximation formula. Linear phase FIR filters are used to
design (to obtain the co efficient theoretically) the decimator. Practically, the desired filter co efficient are
obtained using MATLAB and FDA tool. Filter response is analyzed in both cases. The frequency responses
of the filter are as shown in Figure 1, Figure 2 and Figure 3.

Table 3: Sampling rate converters

Parameter Decimator Interpolator
Design Method Hamming Kaiser’s Optimal Hanning Kaiser’s Optimal
Window Window Method Window Window Method
Method Method Method Method
Order of the filter 481 427 395 360 247 235
Number of 28.8 25.62 23.7 3.24 2.2 2.05
multiplications
( in terms of 106 )
Storage elements 481 427 395 360 247 235

We note that order of the filter is inversely proportional to the frequency deviation Δf. It is also observed, the
required orders of the filter for decimators and interpolators are low in case of equiripple design method.
Hence, it is proved as optimal methods of FIR design results lesser order filter for sampling rate converters
as compared to other FIR designs. This in turn minimizes the delay elements and the storage elements during
the implementation, hence linear phase FIR filters designed by optimal methods are preferred for the
decimators and interpolators as Coefficient symmetry may be exploited and the number of multiplications
per second, storage elements are also minimized during the implementation.

However, the filters for interpolation and decimation always need not to be low-pass filters, which then
invoke the filter design problem to make use of other filters for some applications. In practise, filters like half
band filters and multirate complementary filters are preferred for such applications.

650
 Figure 1. Interpolation Filter Response from Optimal Method

Figure 2. Interpolation Filter Response from Kasier Window Method

Figure 3. Interpolation Filter Response from Hamming Window Method

In practical FIR filters, the magnitude will not be constant in the pass band. The antialiasing filter of this
particular examples have negligible amount of ripples in the pass band. Similarly, there will be ripples in
stop band. From the responses, it is clear that the filter designed by the optimal method closely approximates
the requirements and can be used to design the sampling rate converters. Due to the small transition band,
signal filtering is accurate in the case of FIR filter designed by the optimal method. Whereas window method
results in wider transition band with equal frequencies on both the sides of ideal cut off frequency.

The advantages of window technique are its simplicity and relatively easy to use in determining the filter
coefficients. The set of predefined equations to determine coefficient have made window method to pay
attention for the beginners. But equiripple methods of filter design provides control over the pass band and
stop band thereby controls the frequency response. Among the different window techniques, Kaiser Window
has parameter β that allows adjustment of the compromising in the ripples. By the optimal method, it is
possible to obtain the controlled filter response which is of great interest in audio processing applications.

4. CONCLUSION

In multirate systems, Anti-aliasing and anti-imaging filters are employed to suppress aliasing effect and
remove imaging effect of the converters. Here, the design aspects of multirate filters, their merits and
demerits are discussed. Appropriate filter and design method can be adopted for multirate signal processing
depending upon the application and the required response, Symmetric Linear FIR filters designed by optimal
method are used for sampling rate converters because of their advantages. For multirate systems, optimal
method results less order FIR filters and hence minimizes the storage elements and number of multiplications.
Referring to the Table 2, we believe that the reduction in the order of filter reduces the complexity involved
in the implementation of sampling rate converters and hence many of the multirate systems employ optimal
FIR filter design methods.

REFERENCES
[1] Ronald A Crochiere and Lawrence A Rabiner, “Interpolation and Decimation of digital signals- A
tutorial Review”, Proceedings of the IEEE, Vol 69, No 3, 0018-9219/81/0300-03, March 1981.

651
[2] Prof. Gopal S.Gawande, “Performance analysis of FIR Digital Filter Design Techniques”, International
Journal of Computing and Corporate Research, Vol. 2, Issue 1 January 2012.
[3] Gopal S. Gawande, Bhavna R. Pawar, Dr. K. B. Khanchandani, “Performance evaluation of efficient
Structure for FIR decimation filters using Polyphase decomposition technique”, International Journal of
Electronics & Communication Engineering & Technology (IJECET). Vol. 6, Issue 5, May 2015, pp. 01-
08.
[4] John G. Praokis and Dimitris G. Manolakis, “Digital Signal Processing: Principles, Algorithms and
Apllications”, 4th Edition, Pearson education, ISBN: 9780131873742.
[5] Ronald A Crochiere and Lawrence A Rabiner, “ Multirate digital Signal Processing”, Prentice Hall Inc.,
Englewood Cliffs, ISBN-0136051626.
[6] Priti R. Chandak, Vipul P.Giradkar, Amol T.Wadmalwar , “Design of FIR filter using Matlab Simulink
and Xilinx system generator”, International Journal Of Engineering And Computer Science, Volume 4,
Issue 2, February 2015, Page No. 10384-10387.
[7] Abhishek Kumar, Suneel Yadav, and Neetesh Purohit, “Generalized Rational Sampling Rate Conversion
Polyphase FIR Filter”, IEEE Signal Processing Letters, Vol. 24, No. 11, November 2017.
[8] L. Milic,“Multirate Filtering for Digital Signal Processing: MATLAB Applications” Publisher:
Information Science Reference; 1st edition December 26, 2008
[9] Djordje Babic, and Markku Renfors, “Power Efficient Structure for Conversion Between Arbitrary
Sampling Rates”, IEEE signal processing letters, Vol. 12, No. 1, January 2005
[10] Andreas Franck, “Arbitrary sample rate conversion with resampling filters optimized for combination
with oversampling”, IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, 16
to 19 October 2011.
[11] Qingfeng Jing, Yujia li and Jincheng Tong, “Peformance analysis of multi-rate signal processing digital
filters on FPGA ”, EURASIP on Wireless Communication and Networking, 2019
[12] Mahamudul Hassan, Sheikh Md. Rabiul Islam, Nazifa Tabassumand Xu Huang “Design and
implementation of sampling rate conversion system for Electroencephalogram (EEG) on FPGA device”,
International Journal of Electronics and Communication Engineering (IJECE) Vol. 7, Issue 2, February
- March 2018.
[13] Ravi Kadlimatti and Adly T. Fam, “Efficient Multirate Filter Bank Structure with Full Spectral
Utilization for Multicarrier Communications”, Milcom 2018 Track 1 - Waveforms and Signal
Processing- IEEE 2018.

652