|
Technology
The technology behind algorithms provided by MIKET DSP Solutions is based on the following proprietary technological advances in the area of practical Digital Signal Processing:
- Sub-Optimal Adaptive Signal Processing.
The essence of the adaptive control is NOT to tinker with so-called NLMS equations (wrongly attributed) on an amateur level. The proper approach to adaptive control requires full understanding of the underlying theory and its limitations. You have to account for all real-world non-idealities like weak non-linearity, limited spectrums, singular signals, etc. You have to be fluent with stochastic co-variation analysis of estimation errors and its practical implications. You have to work out a sub-optimal in Calman / Wiener sense strategy for tuning of step size. You have to work out sub-optimal approximation of RLS for each particular application. You have to fully understand how adaptive control works and thus how to stress and load your adaptive application to fully test its capabilities. See the article, which touches on the surfice of the discussed problems.
- Multi-Rate and Sub-Band Signal Processing.
Many of the problems of adaptive control and signal detection may be greatly simplified if formulated using recent advances in multi-rate SP and sub-band bi-orthogonal perfect reconstruction filter-banks. There are many reasons for simplification and many ways to explore those theoretical advances.
An example of using multi-rate analysis for signal detection can be found in the cover story for March'02 issue of the TI's Embedded Edge Magazine. Another example is shown on the picture on the left: transform-domain representation of the room acoustic echo path.
- Sub-Optimal Short Term Spectral Analysis.
The Maximum Likelihood Estimator (MLE) is known to be theoretically optimal (in complex domain) method of spectral analysis. It allows obtaining of the best possible frequency estimation for the given tonal signal. However, although this method requires infinite number of iterations to converge, in a few quite practical cases it may be projected into another domain where sub-optimal results may be achieved in non-iterative way with robust computational scheme.
This approach is used in algorithms like DTMF detector.
The same technology base was used for the novel algorithms of 3D image reconstruction (articles to be published).
|
Papers
|
