Multilayer Perceptrons and Event Classification with data from CODEC using Scilab and Weka
For my first blog, I thought I would introduce the reader to Scilab [1] and Weka [2]. In order to illustrate how they work, I will put together a script in Scilab that will sample using the microphone and CODEC on your PC and save the waveform as a CSV file.
Maximum Likelihood Estimation
Any observation has some degree of noise content that makes our observations uncertain. When we try to make conclusions based on noisy observations, we have to separate the dynamics of a signal from noise.
Implementing Simultaneous Digital Differentiation, Hilbert Transformation, and Half-Band Filtering
Recently I've been thinking about digital differentiator and Hilbert transformer implementations and I've developed a processing scheme that may be of interest to the readers here on dsprelated.com.
Multimedia Processing with FFMPEG
FFMPEG is a set of libraries and a command line tool for encoding and decoding audio and video in many different formats. It is a free software project for manipulating/processing multimedia data. Many open source media players are based on FFMPEG libraries.
Approximating the area of a chirp by fitting a polynomial
Once in a while we need to estimate the area of a dataset in which we are interested. This area could give us, for example, force (mass vs acceleration) or electric power (electric current vs charge).
Deconvolution by least squares (Using the power of linear algebra in signal processing).
When we deal with our normal discrete signal processing operations, like FIR/IIR filtering, convolution, filter design, etc. we normally think of the signals as a constant stream of numbers that we put in a sequence
Roll Your Own Differentiation Filters
Practical guide to constructing differentiation filters from sampled signals using interpolation rather than messy Taylor expansions. It shows how Lagrange polynomials produce forward, backward and central derivative formulas, and how the pseudospectral differentiation matrix D = X'X^{-1} maps sample vectors to derivative estimates. Includes a compact MATLAB snippet and a discussion of node-choice tradeoffs and ill-conditioning for large N.
Helping New Bloggers to Break the Ice: A New Ipad Pro for the Author with the Best Article!
Breaking the ice can be tough. Over the years, many individuals have asked to be given access to the blogging interface only to never post an article.
GPS - some terminology!
GPS looks simple on the surface, but Vivek's post breaks out the core terminology behind how a receiver actually locks on and figures out where it is. Using a bar-room analogy, he maps acquisition, tracking, ephemeris, and almanac to the steps a GPS receiver follows before solving for position from satellite signals.
Welcoming MANY New Bloggers!
A big influx of new voices just joined DSPRelated, and Stephane Boucher introduces the growing roster of contributors and their backgrounds. The post lists dozens of newly approved bloggers, highlights the range of DSP and embedded expertise they bring, and asks readers to leave constructive feedback on posts. It also explains why some applicants may not have been accepted yet and how to apply properly.
Going back to Germany!
A conference conversation turned into a return trip to Germany for Stephane Boucher, this time to visit SEGGER’s headquarters in Dusseldorf and produce videos. The post shares how a chance introduction at ESC Boston led to the invitation, and it teases coverage from SEGGER’s 25th anniversary celebration. He also invites local tips and customer questions before the trip.
A Two Bin Exact Frequency Formula for a Pure Complex Tone in a DFT
Cedron Dawg derives an exact two-bin frequency formula for a pure complex tone in the DFT, eliminating amplitude and phase to isolate frequency via a complex quotient and the complex logarithm. He presents an adjacent-bin simplification that replaces a complex multiply with a bin offset plus an atan2 angle, and discusses integer-frequency handling and aliasing. C source and numerical examples show the formula working in practice.
Signal Processing Summit - Cancellation Policy
The post announces a flexible cancellation policy for the inaugural Signal Processing Summit, an intimate DSP event limited to 70 seats and scheduled in Silicon Valley this October. It explains refundable options designed to give attendees confidence when registering early: a full refund minus a $95 processing fee for cancellations before the end of September, a 50% refund for cancellations in October before October 6, and no refunds after that date. The policy is positioned to help prospective attendees lock in the Early Bird rate, secure discounted hotel accommodations, and plan travel with reduced risk. The announcement frames the policy as a way to remove barriers to commitment and encourages readers who have been undecided to register now and attend the Summit.
A Narrow Bandpass Filter in Octave or Matlab
Building very narrow FIR bandpass filters at high sample rates often yields extremely long impulse responses. This post shows a practical Octave/Matlab implementation that uses complex downconversion to baseband plus a multistage Matrix IFIR and running-sum cascade to slash computation. With the provided example (48 kHz, 850 Hz center, 10 Hz passband) you get <1 dB ripple and >60 dB stopband while running 20x to 100x faster than a single-stage FIR.
Multiplying Two Binary Numbers
Ancient math gives a modern trick for integer multiplication that uses only shifts, parity checks, and additions. Rick Lyons demonstrates the Russian peasant method, shows why it maps to binary right shifts and least-significant-bit tests, and supplies a MATLAB snippet to run the loop. The post also points out a practical tip: put the smaller operand in the halving register to reduce iterations.
Fitting a Damped Sine Wave
Detlef Amberg presents a simple linear-algebra approach to recover frequency, phase, amplitude, and damping of a sampled damped sine wave. Instead of nonlinear fitting, the method casts the waveform as a second-order difference equation, uses linear regression to estimate b and omega, and recovers amplitude and phase by mixing with quadrature carriers; amplitude and damping are then fine-tuned with a gradient iteration. MATLAB code is available on File Exchange.
Matlab Code to Synthesize Multiplierless FIR Filters
Learn how to build multiplierless FIR lowpass filters in Matlab using Canonic Signed-Digit coefficients. The post explains converting Parks-McClellan floating-point taps to scaled integers, then to exact CSD digits, and includes two m-files that search maintap scaling to minimize signed digits while preserving the filter response. Practical notes cover external gain compensation, the 2/3 full-scale CSD limit, and sensitivity to pass/stop edges.
New Papers / Theses Section
Stephane Boucher launched a Papers & Theses section on DSPRelated to gather DSP dissertations and papers in one spot. Authors can submit already-hosted documents or upload PDFs for optional hosting, provided they have sharing rights, and help is available for PDF conversion. Listing your work boosts visibility and opportunities, and non-English documents are welcomed while the section is in beta.
Design of an anti-aliasing filter for a DAC
If you need a practical way to design an anti-aliasing filter for a DAC, this post delivers an Octave/Matlab script that numerically optimizes a Laplace-domain transfer function for linear phase and arbitrary magnitude. The routine models the DAC sample-and-hold sinc response, compensates group delay automatically, and can include an optional multiplierless FIR equalizer. An example shows a 5.4 dB objective improvement and reduced analog Q for easier implementation.
Coupled-Form 2nd-Order IIR Resonators: A Contradiction Resolved
Rick Lyons resolves a long-standing confusion about the coupled-form 2nd-order IIR resonator by deriving its correct z-domain transfer function and explaining why textbooks can appear to contradict pole plots. He shows that with infinite precision the coupled and standard denominators match, but finite-bit quantization of rcos(Θ) and rsin(Θ) changes the z^-2 coefficient and shifts pole positions. Read to learn the correct H(z) to predict quantized behavior and when the coupled form outperforms the standard design.
Controlling a DSP Network's Gain: A Note For DSP Beginners
Rick Lyons calls out a simple but costly mistake beginners make when normalizing digital networks, scaling the input instead of the output. Using fixed-point examples he shows that pre-multiplying an A/D output by 1/8 throws away bits and costs about 18 dB of SQNR. The practical guidance is to place gain control as the final multiplication stage and beware a faulty Simpson's 1/3 integrator example.
Third-Order Distortion of a Digitally-Modulated Signal
Amplifier third-order distortion is a common limiter in RF and communications chains, and Neil Robertson walks through why it matters using hands-on MATLAB simulations. He shows how a cubic nonlinearity creates IMD3 tones, causes spectral regrowth and degrades QAM constellations, and gives practical notes on estimating k3, computing ACPR from PSDs, and sampling considerations.
A DSP Quiz Question
A short visual puzzle from Rick Lyons shows how a common plotting trick can fool even experienced DSP engineers. He presents a 3D circular plot that looks like a triangular window but is actually a 32-point hann window, then explains why the circular projection distorts the view. The post highlights the importance of checking equations and 2D plots before naming a window by sight.
An Efficient Full-Band Sliding DFT Spectrum Analyzer
Rick Lyons shows two compact sliding DFT networks that compute the 0th bin and all positive-frequency outputs for even and odd N, running sample-by-sample on real input streams. The designs reduce computational workload versus a prior observer-based sliding DFT by using fewer parallel paths, while remaining guaranteed stable and avoiding the traditional comb delay-line. A simple initialization and streaming procedure makes them practical for real-time spectrum analysis.
Feedback Controllers - Making Hardware with Firmware. Part 5. Some FPGA Aspects.
This installment digs into practical FPGA choices and board-level issues for a low-latency, floating-point feedback controller. It compares a Cyclone V implementation against an older SHARC-based design, quantifies the tradeoff between raw DSP resources and cycle latency, and calls out Gotchas found on the BeMicro CV A9 evaluation card. Engineers get concrete prompts for where to optimize: clocking, DSP-block use, I/O standards, and algorithm partitioning.
Above-Average Smoothing of Impulsive Noise
This post introduces a smoothing trick that behaves a lot like a moving average for high-frequency noise, but does a much better job of suppressing impulsive spikes. Rick Lyons shows how the corrected average is computed from the sample count, the sample imbalance around the mean, and the total deviation. He also compares the method against a standard moving average on a noisy step signal, where the improvement is easy to see.
Generating Partially Correlated Random Variables
Designing signals to match a target covariance is simpler than it sounds. This post shows how to build partially correlated complex signals by hand for the two-signal case, then generalizes to N signals using the Cholesky decomposition. Short MATLAB examples demonstrate the two-line implementation and the article highlights numerical caveats when a covariance is only positive semidefinite.
Off-Topic: A Fluidic Model of the Universe
Cedron Dawg develops a Newtonian, fluidic model where space is a compressible "fluff" and particle motion is governed by a simple refractive steering equation. He shows how rho = ln n links index, permittivity and permeability to a gravity-like potential, derives a massive-particle steering law, and works through orbit and disk solutions that produce MOND-like effects while conflicting with General Relativity. The paper highlights concrete formulas and numerics to test the hypothesis.
Hidden Linear Algebra in DSP
Linear algebra is hiding in plain sight inside many DSP techniques, not just abstract theory. By treating linear systems as matrix operators y = A x you reveal Toeplitz structure in LTI systems, connect to covariance matrices, and gain geometric intuition via eigenvalues and eigenvectors. This matrix viewpoint complements convolution-based thinking and offers practical tools for filter and channel analysis.
Reducing IIR Filter Computational Workload
Rick Lyons demonstrates a simple, practical way to cut the multiply count for IIR lowpass and highpass filters by converting them into dual-path allpass structures. The conversion preserves the original magnitude response while drastically reducing multiplies per input sample, for example turning a 5th-order IIR that needs 11 multiplies into an equivalent allpass form needing only five. The linked PDF includes theory, implementation notes, a design example, and MATLAB code.
