I know that papers these days build up on a lot of literature and hence are very compactly written. There is also the issue of page limits. I've heard that papers are written for those who are already familiar with the particular area.
I know that completely understanding a paper takes time and effort but I have come across papers that are easy to follow and also easy to reproduce in MATLAB. Such papers give the beginner level researcher some confidence. A good example would be: https://ieeexplore.ieee.org/document/508168
May be some of the readers don't have IEEE access. But to me this was the most appropriate example.
Now what I wanna know is what papers have you come across that seem very easy to follow (and reproduce/simulate) and it' s almost like reading a book?
The areas I am referring to are
1) Detection and Estimation [my favorite subject in grad school (MS EE) ]
2) Signal processing for communications
May be such papers will be older but of course latest publications would be preferable. Most likely such papers are IEEE Signal processing letters, IEEE communication letters or some conference papers.
P.S. A couple of things:
a) I posted this question on dsp stackexchange and someone suggest dsprelated.com's forums would be a more appropriate place for this.
b) The definition of easy is relative so to give the reader some perspective: I have a Master's in Electrical Engineering (MS-EE) specializing in Digital Systems and Signal Processing. I have just one publication in the area of estimation theory. It can be found here: https://ieeexplore.ieee.org/document/8675332
On the topic of general DSP there used to be two magazines that published, as you say, "papers that are easy to follow and also easy to reproduce in MATLAB." Those magazines were: 'Embedded Systems Programming' magazine and Personal Engineering magazine. Sadly, both of the magazine publishers are gone (no longer in business).
When it comes to IEEE-published papers you will see VERY VERY few "papers that are easy to follow and also easy to reproduce in MATLAB." There are three reasons for this sad situation:
(1) The IEEE publications' editors, over the years, have developed editorial standards that inhibit "easy to follow papers." Years ago I read the IEEE Signal Processing Society's "Guidelines for Authors" instructions. And their Rule# 1 was: "Do not explain the obvious." I am not joking. I suppose they have that rule to reduce the page count of their published papers, but I believe such a rule inhibits the communication of ideas. Such a rule encourages authors to falsely assume that their readers know everything that the author knows.
(2) Typical IEEE authors skip over necessary technical explanations by referring to previously written papers. They present a equation and figuratively say, "Here's an important equation. If you want to know it's origin, derivation, and meaning, see References [X, Y, & Z]." This practice reduces a paper's page count but makes life miserable for the reader.
(3) Typical IEEE authors are VERY smart people, with strong analytic skills, powerful abstract thinking ability —- people who are very fluent in the language of advanced mathematics. However, typical IEEE authors are too often not skilled at writing, and the typical magazine editor is not skilled enough to recognize, and correct, poor writing. Typical IEEE authors often violate the fundamental rules of clear, understandable, writing. The authors too often unconsciously assume their audience can read the author's mind.
OK enough of my ranting and raving here. In closing I must say that in 2003 the IEEE Signal Processing Magazine (SPM) started publishing a column called "DSP Tips & Tricks" which presented "readable" articles on a variety of DSP topics. If you've paid your money and have access to the SPM then you can go through the magazine's archives and view those DSP Tips & Tricks articles.
By the way szak1592, on page 106 of the recent May 2019 issue of the IEEE Signal Processing Magazine is an article titled: "Reproducible Research: Best Practices and Potential Misuse." That thought-provoking article describes a fundamental shortcoming in modern-day signal processing publications.
So basically unless you have a PhD plus a few years of post PhD experience, you are doomed.
Maybe the rules for authors that Rick mentions should be modified as follows:
After you are done writing the paper add 100 or 200 words to explain "the obvious".
Comparing the effort/cost involved in explaining the "obvious" versus the effort/time spent by new researchers in understanding the "obvious", I think authors should mention the obvious.
Well, ...we're not totally doomed. What we poor readers must typically do when plowing through a signal processing paper, as I wrote in the Preface of my book, is "figuratively grab a pick and shovel, put on a miner's helmet, and try to dig the useful information out of a mountain of mathematical expressions."
I was happy you posted a link to the "frequency estimation" paper by Kim, et al. Because you said that paper was easy to read and understand I decided to see if I could understand it. On a scale of 1 -to- 10 (with 10 being a VERY readable paper) I give the Kim paper a rating of 6. My rather poor rating of that paper is based on the following:
 In their Eq. (8) they give an equation for estimating the frequency of a complex-sinusidal signal. In that equation is the variable w_sub_k which the authors made no attempt to define. If the readers are unable to pray to their God to find out what is w_sub_k, then that's the reader's problem.
 In my opinion the algebraic "black magic" in Eq. (15) up through Eq. (23) deserves a much more thorough explanation from the authors.
 The authors give their so-called improved equation for estimating the frequency of a complex-sinusidal signal, Eq. (23), that also contains the undefined variable w_sub_k. Because w_sub_k is undefined, based on the text in this paper the reader has zero chance of repeating the authors' results or implementing the authors' "improved" freq estimation algorithm. When I was an editor for the IEEE Sig. Proc. Magazine I would have never allowed this situation to happen in a published article.
szak1592, I'm going try to find a copy of some of the above paper's references in an attempt to understand what in the "H E double-hockey-sticks" is the meaning of that mysterious undefined variable w_sub_k. Shame on the papers' authors and editor for making me do this.
Hi Rick Lyons,
Yes, this w_sub_k is undefined. Actually you have to go through all the steps, meaning that for a different K (the authors showed the derivation for K = 2) you have to follow the steps and derive the estimator (which I was able to do) and then you can simulate using the equations that you derive.
In going through the steps, you will end up with a different form of eq. 15 and eq. 19 and then you can plug them into eqs. 20 and 21. So the procedure is not that generic, you have to derive it every time K changes.
I remember seeing a paper that defines this w_k. I'm guessing it was a review paper by Mark Fowler. But I agree with you, esp. point no. 2 about the algebraic black magic.
BTW, since you rated that paper 6 on the readability/reproducible scale, could you please mention some of the more readable ones? I'm just interested in learning and digging deeper in the world of signal processing.
Asking me to recommend "readable" DSP articles is like asking me to recommend good Rock-n-Roll songs. It's all a "matter of opinion." And my opinion is no more valid than the next guy's opinion.
In any case, in the IEEE Signal Processing Magazine here's a list of articles whose "readability" is higher than average:
Jan 2003 Lyons, R. "Interpolated narrowband lowpass FIR filters"
Mar 2003 Jacobsen, E.; Lyons R. "The sliding DFT"
May 2003 Turner, C. "Recursive discrete-time sinusoidal oscillators"
JUL 2003 Hars, L. "Frequency response compensation with DSP"
Sep 2003 Donadio, M. "Lost knowledge refound: sharpened FIR filters"
Nov 2003 Kotteri, K.; Bell, A.; Carletta, J. "Quantized FIR filter design using compensating zeros"
JAN 2004 Lyons, R. "Another contender in the arctangent race"
Jan 2004 Jacobsen, E.; Lyons R. "An Update to the sliding DFT"
MAY 2004 Lyons, R.; Bell, A. "The swiss army knife of digital networks"
JUL 2004 Cordesses, L. "Direct digital synthesis: a tool for periodic wave generation (part 1)"
SEP 2004 Cordesses, L. "Direct digital synthesis: a tool for periodic wave generation (part 2)"
NOV 2004 Varma, K.; Bell, A. "JPEG2000 -choices and tradeoffs for encoders"
JAN 2005 Storn, R. "Designing nonstandard filters with differential evolution"
MAR 2005 Allie, M.; Lyons, R. "A root of less evil"
MAY 2005 Shiung, D.; Huei-Wen F.; Lyons, R. "Filtering tricks for FSK demodulation"
JUL 2005 Losada, R.; Pellisier, V. "Designing IIR filters with a given 3-dB point"
SEP 2005 Farrell, D.; Oakley, A.; Lyons, R. "Discrete-time quadrature FM detection"
NOV 2005 Rader, C. "Generating rectangular coordinates in polar coordinate order"
JAN 2006 McNames, J. "An Effective Color Scale for Simultaneous Color and Gray-scale Publications"
MAR 2006 Borgerding, M. "Turning Overlap-save Into a Multiband, Mixing, Downsampling Filter Bank"
MAY 2006 Rajan, S.; Sichun Wang, S.; Inkol, R.; Joyal, A. "Efficient Approximations For the Arctangent Function"
JUL 2006 Losada, R.; Lyons, R. "Reducing CIC Filter Complexity"
SEP 2006 Ylostalo, J. "Function Approximation Using Polynomials"
NOV 2006 Engelberg, S. "Implementing a Sigma Delta DAC in Fixed Point Arithmetic"
JAN 2007 Berchin, G. "Precise Filter Design"
MAR 2007 Vassilevsky, V.L. "Efficient Multitone Detection"
MAY 2007 Jacobsen, E.; Kootsookos, P. "Fast, Accurate Frequency Estimators"
JUL 2007 Harris, F. "Ultra Low Phase Noise DSP Oscillator"
SEP 2007 Lyons, R."Turbocharging Interpolated FIR Filters"
NOV 2007 Bariska, A. "Recovering Periodically Spaced Missing Samples"
MAR 2008 Yates, R. and Lyons, R. "DC Blocker Algorithms"
JUL 2008 Barker, D. "Efficient Resampling Implementations"
SEP 2008 Engelberg, S. "Precise Variable-Q filter Design"
NOV 2008 Turner, C. "Slope Filtering: An FIR Approach to Linear Regression"
MAR 2009 Lyons, R. "Improved Narrowband Low-Pass IIR Filters in Fixed-Point Systems"
MAY 2009 Givens, M. "Enhanced-Convergence Normalized LMS Algorithm"
JUL 2009 Turner, C. "An Efficient Analytic Signal Generator"
NOV 2009 Tan, L.; Jiang, J. "Novel Adaptive IIR Filter For Frequency Estimation and Tracking"
MAR 2010 Yoon, T. and Joo, E. "A Flexible Window Function For Spectral Analysis"
JUL 2010 Shen, Z. "Improving FIR Filter Coefficient Precision"
SEP 2010 Turner, C. "A Fast Binary Logarithm Algorithm"
NOV 2010 Duda, K. "Accurate, Guaranteed-Stable, Sliding DFT"
MAR 2011 Lyons, R. "Reducing FFT Scalloping Loss Errors Without Multiplication"
MAY 2011 Bi, G. and Mitra, K. "Sampling Rate Conversion in the Frequency Domain"
JUL 2011 Auger, F.; Zhen, L.; Feuvrie, B.; and Li, F. "Multiplier-free Divide, Square Root, and Log Algorithms"
SEP 2011 Hongwei Guo; "A Simple Algorithm for Fitting a Gaussian Function"
NOV 2011 Abhishek Seth and Woon-Seng Gan "Fixed-point Square Roots Using L-bit Truncation"
JULY 2017 Lyons, R. "Digital Envelope Detection: The Good, the Bad, and the Ugly",
NOV 2017 Lyons, R. "Fast and Low-Complexity Atan2(a,b) Approximation"
Thanks Rick. Hopefully I'll find some, if not all, of these articles to be readable.
Oh, a veritable gold mine! Interestingly enough, Matt Donadio, author of the Tips and Tricks article on sharpening filters (really cool trick!) worked together a couple decades ago. Small world!
Thanks, Rick! This list is a keeper!
>When it comes to IEEE-published papers you will see VERY VERY few "papers that are easy to follow and also easy to reproduce in MATLAB."
I think that there are several factors at play here:
1. For science to "work well", we need the experimenters that find novel mechanisms. We need the theoreticians that offer predictions. We need the people that confirm other peoples results. We need reviewers. But we also need people writing review articles, people consolidating the cutting edge work into one coherent package. All of these are needed, but I think that current scientific career options are heavily biased towards a subset.
2. Science must be reproducible. For tasks that are usually "easy" to reproduce, I suggest that an author is barred from publishing a plot or a table unless she also offer a csv file of datapoints in the rawest possible form (sensor data, speech samples, listening test responses...) and some code (Python, Excel,...) leading to those plots. The code need not be particularly efficient, readable, generic,... but it should work as suggested by the text in the paper for at least the input leading to the conclusion-forming graphs/tables.
3. People may not be only an academic. They can also be an entrepreneur, an industry employee etc. Thus, they can have a vested interest in not revealing crucial details (such as how to regularize a formula in the presence of noise or finite precision).
4. The truly great minds (often) are capable of expressing complex ideas in a comprehensible manner. Most of us aren't truly great minds. When you know some math, did not get particularly novel results, or did not really comprehend your line of study, you can always wrap your results in a veil of scary math. Many reviewers may simple assume that you know what you are doing.
5. The world is changing. People are gravitating towards social media and fast, personalized newsfeeds rather than curated printed news. Scientists and engineers are people, too. I see people skimming through a large number of (non-peer-reviewed) arxiv articles that are pushed at large numbers and short latency. Some tasks (machine learning) lends itself well to the actor with the largest dataset and the most compute power (typically hyper scalers), i.e. a small group of bright academics might have a hard time keeping up with a large industry actor. This will probably change the motivation and behavior of publishing scientists.
No problem with the rant re: IEEE Signal Processing magazine, my least favorite IEEE magazine, at least now that you're not doing the DSP Tips and Tricks anymore :( So much math in the articles, particularly the matrices, that I get totally lost. Although I did feel a bit better about myself a couple years ago at an IEEE talk when I met an SP researcher with a PhD and many years experience who said even he didn't understand the articles in SP magazine!
I contrast SP mag with my favorite IEEE magazine, the Microwave magazine. Now, maybe it's just that I'm more familiar with the continuous time world and playing with the instruments that work in it, but it seems to me that the Microwave mag nails the sweet spot between the mostly-advertising trade rags and the stuffy journals that only people with multiple PhD's can read!
Well, I guess I haven't contributed much worthwhile to the topic. Kind of ranting away myself here! I guess I better dig in and take the Kahn Academy class(es) on linear algebra or something. I was hoping to find some good references here. I haven't gotten to the May issue of SP mag yet, but the Reproducible Research article sounds like a good one to read.
Hi jimelectr. Ah but you did contribute to this topic! I enjoy hearing the opinions of working engineers regarding the literature of electrical engineering and signal processing in particular. And I agree with you that there's an awful lot of matrix algebra sprinkled throughout the IEEE Signal Processing Magazine.
I studied matrix algebra many decades ago while I was in college. But it doesn't take long to forget some mathematical tool you've learned if you do not use it regularly. I imagine matrix algebra is can be useful in solving certain types of mathematical problems, but over my career I never had the need to use matrix algebra to help me in any of my engineering work.
To continue the discussion of linear algebra: the classic application we learned in EE was steady-state simulation of linear RLC circuits. A circuit with N loops can be solved at a given frequency using N simultaneous linear equations. The typical homework problem circuit had 3 loops, to allow solving by hand.
With the advent of Matlab and Mathcad, which allow writing matrix equations symbolically, it became easy to solve bigger circuits.
I think that your question is really nice but in same time there is a huge set of possible answers. However, I will try to offer suggestions related to array signal processing and subspace based methods. These could be really nice papers to read:
- Subspace Based Signal Analysis using Singular Value Decomposition (https://ieeexplore.ieee.org/document/237536)
- Detection and estimation in sensor arrays using weighted subspace fitting (https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=97999)
More modern approach to array SP and nice overview of the related topic can be found in this book chapter:
- Sparse methods for direction-of-arrival estimation (https://www.sciencedirect.com/science/article/pii/B9780128118870000110)