Multiple Peak Count Analysis, MPCA

In underwater applications such as current measurements using Doppler profiling systems, it is of great interest to study the current versus depth. These acoustic systems operate using a short “ping”, that consist of a sinusoidal burst that is transmitted into the water by a piezo-ceramic transceiver. By listening to the change in Doppler frequency in the backscattering signal, created by entrained air bubbles and other small particles, it is possible to calculate the mean current versus time. However, there could be turbulence in the water creating simultaneous currents and layers could change the current direction rapidly, and this could have a negative impact on the current calculation algorithms. These multiple current phenomenon has also been difficult to analyze due to the bandwidth-time limitation. The transmitted pulse is short, and the depth where the current should be known is also typically small. When using this new non-linear filtering method, Multiple Peak Count Analysis – MPCA, these current change effects has become very visible, and the method shows superior performance as compared to classical methods. The MPCA method has several similarities to the Symmiktos Method  that has been developed by Dr Thomas Lagö, which is a non-linear estimation method for the Doppler frequency. When using a classical approach to a 3D spectral plot of the data, there is direct support for that in Matlab® using the specgram command. In this case, there is a particular interested in finding multiple peaks in the data, and by pre-processing the data this information can be enhanced. A non-linear pre-filtering method has thus been developed where the peaks are extracted in the spectra and small peaks truncated. If the analyzed signal is random, the standard deviation is 100% for each estimate. However, the FFT will produce the same result every time, given that the time signal xr(t) is deterministic. A short-time FFT has been used as a calculation of the spectral content, where a small time data block is processed using a much larger time analysis block size. By doing this, an interpolation in the frequency domain is achieved, and thus, the frequency lines are not as sparsely sampled as when using an FFT without zero-padding, and this will making the frequency spectra less cornered. Zero-padding only increases the frequency domain sampling rate, but does not increase the actual frequency resolution. The left figure represent an advanced ST-FFT and the right is the MPCA method. The result speaks for itself in the pictures below. Dr Thomas Lagö has presented multiple papers on the topic.

MFPT Medical