Waveform Information Vector Technology

WIV vector processing is a unique time-domain event stream processor whose capabilities are derived from its ability to closely model the human cochlea.  WIV’s stream processor possesses minimal calculation overhead and is well suited for high performance, real-time applications.  As a cochlear analog, WIV inherits many of the ear’s ability – detecting and differentiating multiple frequencies, spatially separating multi-tone signals to determine the direction of arrivals, and extracting sufficient information from waveform data to facilitate speech processing. Coded in C, C++ and Objective C, it is extremely fast and runs in real-time on microprocessor chips.

Technical Summary

WIV’s methodology evolved from research and development of a passive non-scanning airborne electronic reconnaissance systems.  Its purpose was to uniquely identify multiple radar sources and determine their direction of arrival.  The solution was a unique time-based method to extract waveform information at the moment of a zero-crossing (real and complex).  An approach used in the development of the AN/ASQ-18.  Research continued in the private sector culminated in the creation of the WIV waveform analysis process. Follow-up research, focused on the development of a software version, was accomplished in 1986.  Subsequent research concentrated on speech recognition applications until 2008 when research and development were transferred to WIV, LLC. This new organization then tasked itself with seeking opportunities to create and improve customer applications.

 WIV Processor

The inherent nature of a Waveform Information Vector (WIV) is to quantify cyclic patterns into a sequence of event particles that embed waveform time, space, and energy information.  At each zero crossing (both real and complex), a particle is created with a precise crossing timestamp, signal amplitude, and duty ratio calculated at the time of the crossing (Figure 1).  A downstream processor groups two sequential WIVs – three leading edges – and uses the groups to identify and classify a waveform’s periodicity elements along with its creation timestamp.

Figure 1

To determine periodicities, the WIV groups pass through a tapped shift register located in a Periodicity Sorting Matrix (PSM) where duty cycles are associated with a specific periodicity. (Figure 2) The PSM aligns the duty cycle of the WIV stream with ‘bin’s’ associated with periodicities measured in Hz. In a method similar to a table lookup, the PSM deconstructs the sorted particle into the specific components needed by an application.

Figure 2

The PSM’s ‘bins’ are created as equal divisions of the entire periodicity range under study. The granularity of the PSM determines the periodicities associated with a bin and therefore the output’s resolution. The ‘bins’ hold an incremented count of the number of times a specific periodicity passes through the PSM. The stored counts indicate the relative dominance of the frequencies associated with that ‘bin’.

The entire process has minimal computational needs other than half-wave rectification and the insertion of a logarithmic delay line – both of which are known to exist in the cochlea.  These low computational requirements offer a uniform, consistent processing time along with the ability to process in real-time.

WIV Processing vs. Other Methods

While WIV’s event processor is unique and has the ability to produce the same, yet more precise results as those derived from other methods such as FFT and Wavelets.  Figure 3 presents a side-by-side comparison of FFT vs. WIV.

In this case, the waveform shown is a low amplitude 220Hz sine wave summed with a high amplitude 440Hz sine wave. Sampled at 44100Hz, the dataset contained 65,536 data points. Both the low amplitude 440 Hz sine wave and the high amplitude 220Hz waveform are clearly seen on the FFT display.

The WIV graph shows a time snapshot of the original signal (top), the amplitude of the peaks alternating from .5V p/p to 1V p/p (middle), and the periodicity (bottom).  In the periodicity graph, Colors represent amplitude with red being high amplitude and blue low.

Figure 3

FFT (x=Hz, y=dB)

WIV (Waveform, Amplitude, Periodicity)

In comparison, both the FFT and the WIV time domain event processing detected the same frequencies. Only WIV was able to demonstrate a pulse-by-pulse detailed real-time output.  While capable of resolving the same periodicity data, the WIV processor offers many advantages.

  • Although defined as time-domain, WIV processing doesn’t use computationally intensive algorithms as needed by applications such an FFT, Wavelets, or time-domain methods.
  • WIV is a unique approach, as waveform information is not degraded in processing. The WIV particle retains sufficient information to permit the reconstruction of the original waveform.
  • WIV’s computational speed is typically defined by the CPU speed and not the sampling rate.
  • WIV’s functional frequency range is 10Hz to 100,000 Hz and higher.
  • WIV has high noise rejection and easy bandwidth filtering.
  • WIV maintains precise waveform timing for applications needing exact signal arrival time.
  • WIV’s small code size and low computational needs support applications running on low power microprocessors.

WIV Applications

The WIV processor, written in C, C++ and Objective C, is used as a DLL in the Windows operating system.  The WIV processor’s output is customizable to the application’s needs by modifying the sorting matrix to vary its output dataset.  Outputs vary from simple periodicity data to comprehensive prosody datasets for speech processing.   A matrix of WIV capabilities is shown below (Figure 4).  In the grid, the ‘Demonstrated’ column indicates WIV’s capabilities that have already been demonstrated and/or published. The ‘Supports’ columns identify capabilities consistent with known capabilities of the WIV processor but have not been tested.

Figure 4

Comments are closed.