RMIT University
School of Electrical and Computer
Engineering
EEET2369 – Signals and Systems
Assignment #2
Lecturer: Dr. Katrina Neville
Student Name: Alakawalage Don Bhanuka Sachintha Samarakkody
Student Number: s3490356
Submission Due Date: 15th May 2015s3490356 Alakawalage Don Bhanuka Samarakkody
1
Introduction
The main purpose of this assignment is to investigate the uses of the SMA (Simple Moving
Average) and EMA (Exponential Moving Average) filters and the uses of the filters, the pros
and the cons of using each filter for its uses ,and to identify which filter is better suited for
which purpose and to what extent the data from each filter can be trusted, which filter most
accurately allows trends to be identified and how the two filters together can provide a
unique set of information relevant to making important decisions. Eg: what instances to buy
and sell shares.
How the general difference equations and transfer functions are found and how the SMA
and EMA filters are related to the above equations and the qualities of each of the filters
whether , high pass, low pass , etc. and how that can be identified from the frequency
response graph created by using the freqz() function in MATLAB.s3490356 Alakawalage Don Bhanuka Samarakkody
2
Simple Moving Average
(P1HW-1)
Impulse graph for a filter with a length of 5 samples
(P1HW-2)s3490356 Alakawalage Don Bhanuka Samarakkody
3
The graph in figure (1) below shows the simple moving average filter for the closing stock
prices for Caltex
Figure (2) below shows the same graph as above, but the engaging of the filters have been
taken out .There is an engaging delay of 5 days and of 20 days in the filtered samples
respectively.s3490356 Alakawalage Don Bhanuka Samarakkody
4s3490356 Alakawalage Don Bhanuka Samarakkody
5
Discuss:
1. How well do these two filters smooth the stock price data?
As clearly seen in figure (2), the short term (5 days) and the long term (20 days) filters both
minimize the effect of high-frequency fluctuations or noise and remove any timely trends in
frequency, The filters have brought a ‘smoothened’ effect into the original data.
However when the two filters the short term and the long term are compared with each
other the long term filter has a higher smoothening effect because a larger amount of
samples or data points have been taken into account, so the longer term filter is more
effective in the identification of trends in the stock price than the short term filter.
2. Compare the delay in the two methods. For both filters, how many initial output
samples/data points should be discarded before a reliable indication of trend can
be obtained from this output? Why is the case?
The longer term filter has a greater delay than the short term filter simply because it uses
the average data from every 20 samples, instead of the 5 which the short term filter uses,
hence it takes a bit longer to adjust to the changes in the stock price.s3490356 Alakawalage Don Bhanuka Samarakkody
6
3. Use the Math lab freqz() function on these two filters. Do they behave as low-pass,
high pass, band-pass or band-stop filters? How does the length of the filter affects
the passband region (frequency that the filter passes)?
By observing the graphs below the, properties of a graph of a low pass filter can be
identified as the band of the phase response keeps increasing slightly every time.
The lower the length of the filter the greater number of frequencies pass through giving a
closer to accurate signal, but the greater the length the less frequencies that pass through ,
which produce a smoother curve.s3490356 Alakawalage Don Bhanuka Samarakkody
7s3490356 Alakawalage Don Bhanuka Samarakkody
8
Exponential Moving Average
(P2HW-1)
(P2HW-2)s3490356 Alakawalage Don Bhanuka Samarakkody
9s3490356 Alakawalage Don Bhanuka Samarakkody
10s3490356 Alakawalage Don Bhanuka Samarakkody
11
Discuss:
1. Compare the behaviour of the EMA filters with the SMA filters from part a.
The SMA is commonly used as it provides a simple solution for problems, The SMA weights
each data point equally and it is calculated by dividing the number of samples by the total
number in the series. But in the EMA more weight is given to the recent data points and
reacts quickly to changes in trends, being favourable for people seeking to identify trends.
2. Using these long and short-term average lines, identify the times when it would
have been good to buy and sell stocks in your company. Justify your answer from
the explanation at:
http://www.asx.com.au/prices/exponential_moving_average.htm
A trading signal is issued when a short term average line crosses over a longer term average
line. A buy signal is produced when the short term average crosses above the longer term
average. A sell signal is produce when the short term average line crosses below the long
term average line.
Two instances to buy and sell are indicated in the graph below.
Time to sell
Time to Buys3490356 Alakawalage Don Bhanuka Samarakkody
12
3. Again use the MATLAB freqz() function on these two filters and examine their
frequency responses.
Both the frequency responses end on an increasing straight line, which indicates that the
next step would be higher, that indicates that the EMA filter acts as a low pass filter as well,
similar to the SMA filter above, but since the equation for the second part ends up with only
a few values for the denominator and the numerator the pattern cannot be clearly
identified as earlier.s3490356 Alakawalage Don Bhanuka Samarakkody
13s3490356 Alakawalage Don Bhanuka Samarakkody
14
Appendix
Some variables in the code below are similar and were commented out when obtaining the
results.s3490356 Alakawalage Don Bhanuka Samarakkody
15s3490356 Alakawalage Don Bhanuka Samarakkody
16s3490356 Alakawalage Don Bhanuka Samarakkody
17
References
1. FIR and transfer functions, pg 1- 4
URL:
http://www.astro.rug.nl/~vdhulst/SignalProcessing/Hoorcolleges/college07.pdf
2. The Scientist and Engineer's Guide to Digital Signal Processing, Chapter 15,Moving
average Filters, pg 278-279,
URL:http://www.analog.com/media/en/technical-documentation/dspbook/dsp_book_Ch15.pdf
3. Digital Filters, FIR Filters , section 6
URL : http://www.analog.com/media/en/training-seminars/designhandbooks/MixedSignal_Sect6.pdf
4. James H. McClellan, Signal Processing First, International ed., Upper Saddle River,
NJ 2003,pg 110-131
5. Dr. Katrina Neville, RMIT University ,City campus , Melbourne, EEET2369 Signals and
Systems :
Lecture 5: Sampling and Aliasing
Lecture 6: FIR Filters
Lecture 7: Frequency Response of FIR Filters
Lecture 8: The z-transform
6. MATLAB documentation
Oppenheim, Alan V., Ronald W. Schafer, and John R. Buck. Discrete-Time Signal
Processing. 2nd Ed. Upper Saddle River, NJ: Prentice Hall, 1999.
URL: http://au.mathworks.com/help/signal/ref/freqz.html