|
|
|||||||||||||||||
Course Catalog 2011-2012
SGN-2156 System Level DSP Algorithms, 4 cr |
Additional information
This course contains numerous common sense ideas
which should be of great value for an expert in
the design of digital filters and multirate
filter banks. This course will not be lectured
but it can be passed as an exam. In cases where a
student is interested in the course (the material
is available in the home page of the lecture; in
addition the previous exercise material will be
given on request), there is only a need to
contact the lecturer. He is ready to arrange
extra examinations provided that the student will
talk with the lecturer on those conditions that
are required to pass the course. The lecture
notes should be ready for the self studying,
thereby making it straightforward to pass the course.
Suitable for postgraduate studies
Person responsible
Tapio Saramäki
Lessons
| Study type | P1 | P2 | P3 | P4 | Summer | Implementations | Lecture times and places |
| |
Requirements
Final examination and 2 assignments out of 3.
Principles and baselines related to teaching and learning
-
Learning outcomes
After finalizing the course with a good grade, the goal is to make the student aware of how DSP experts should work in a fluent interactive manner with experts, who concentrate on implementing (as a VLSI circuit or using a signal processor) DSP algorithms for arriving together at good final products. For this purpose, the student should have more motivation, encouragement, and knowledge, among others, on the following aspects: -Why is it crucial to try to partially or totally avoid the use of a general multiplier element in VLSI implementations by first reducing the number of arithmetic operations and then expressing all coefficients values or most of them by few powers-of-two terms in the DSP algorithm at hand? -How to increase the throughput of a signal processor by decreasing the number of arithmetic operations in the DSP algorithm? -In most cases, there are several optional filter or filter bank solutions to meet the same criteria, that is, it is not worth selecting the first solution coming to mind; further brainstorming and intuition, which can be improved by working as a DSP expert for several years, result typically in a considerably better final product. -How to exploit optimization theory being directly available in the MATLAB Optimization Toolbox in order to generate a good overall product which meet the given criteria subject to the constraints stated by the application at hand and the implementation form in use? -How to find and use simple practical tricks, not available in existing textbooks, e.g., for quantizing the coefficients of direct-form FIR filters or IIR filters being implementable either as a cascade-form structure or as a parallel connection of two all-pass filters. -Why is it crucial to generate own thinking tools, instead of using formulas describing the problem, in order to become a real DSP expert? They should also be able to use several general-purpose MATLAB algorithms generated by the lecturer for exemplifying the above topics.
Content
| Content | Core content | Complementary knowledge | Specialist knowledge |
| 1. | The main goal of the course is to learn how DSP experts should work in a fluent interactive manner with experts who concentrate on implementing (as a VLSI circuit or using a signal processor) DSP algorithms in order to end up, by means of a friendly team work, with good final products. | ||
| 2. | How to develop the efficient digital signal processing algorithms on the system level in such a way that (a) the overall throughput of the signal processor implementations increases; (b) in VLSI circuit implementations, both the silicon area and power consumption become smaller as well as the maximal sampling frequency becomes larger? For achieving these goals, it is crucial to implement the DSP algorithm such that (a) for signal processor implementations, the number of arithmetic operations is decreased as much as possible; (b) for VLSI circuits, the use of the general multiplier element(s) are avoided as much as possible by expressing all or most of the coefficient values by few powers-of-two terms. These aspects are exemplified by the part "Design of various kinds of digital filters meeting the same criteria". | ||
| 3. | Design of digital filters using identical building blocks - Straightforward techniques to generate multiplication-free linear-phase FIR filters, IIR filters, and half-band linear-phase FIR filters. | The part "Design of digital filters using identical building blocks" covers three alternatives. Only the basic principle has to be learned. | |
| 4. | Finite word length effects in practice: (a) How to easily quantize the coefficient values of direct-form linear-phase FIR filters?; (b) How to easily quantize the coefficient values of IIR filters, which are implemented as a cascade of second- and first-order blocks?; (c) How to easily quantize the coefficient values of IIR filters, which are implemented as a parallel connection of two allpass filters?; (d) The validity of the noise model, which is commonly used to estimate the output noise due to the multiplication roundoff errors. | ||
| 5. | Some elegant designs based on the use of recursive running sum filters. | There are two journal articles on the part "Some elegant designs based on the use of recursive running sum filters". The key idea of using these filters has to be learned. | |
| 6. | Design of filters and filter banks by optimization: Applications. | The part "Design of filters and filter banks by optimization: Applications" is very long. Again, only the main ideas have to be learned. | |
| 7. | Recent advances in ant research; those students who have actively participated in the lectures badly deserve during the last lecture some fun. |
Evaluation criteria for the course
Course is graded on the basis of answers to exam questions. Very good grade is obtained when exam questions are correctly answered and 2 homeworks of 3 are accepted. Course acceptance threshold approx. is half of the maximum exam points. 3th homework is a volunteer work and is prized with increasing the exam result by one grade provided that the threshold is passed.
Assessment scale:
Numerical evaluation scale (1-5) will be used on the course
Study material
| Type | Name | Author | ISBN | URL | Edition, availability, ... | Examination material | Language |
| Summary of lectures | System Level DSP Algorithms | Tapio Saramäki | English |
Prerequisites
| Course | Mandatory/Advisable | Description |
| SGN-2056 Digital Linear Filtering II | Mandatory | |
| SGN-2106 Multirate Signal Processing | Advisable |
Prerequisite relations (Requires logging in to POP)
Correspondence of content
| Course | Corresponds course | Description |
|
|
|
More precise information per implementation
| Implementation | Description | Methods of instruction | Implementation |