Signal Flow

Block diagram of the signal flow in IAF processing

Fig 1. Block diagram of the signal flow in IAF processing

Mathematical modeling is applied to the input signal and the extracted model parameters are then used to control an adaptive filter. The output of the adaptive filter is fed back to the modeling unit in order to control the modeling adaptation in time. The adaptive filtering produces a signal carrying the information about characteristics of the input source. Finally, the extracted signal characteristics are enhanced to produce the “characterized” output signal.


First the input signal is analyzed. According to the input and the current output of the adaptive filter, the adaptive filter is tuned and adjusted. Majority of the characterizing information lies in the transient components of the input signal rather than in the short-time steady state components. Therefore, the modeling procedure is focused on accurately modeling time-varying properties of the prominent spectral components. The difference is used to define the characteristic information.

Adaptive filtering

General adaptive system

Fig 2. General adaptive system

The purpose of the general adaptive system is to filter the input signal so that it resembles the desired signal. The error signal is the difference between the desired signal and the output signal. This difference is fed back to the adaptive process algorithm that evaluates the resemblance between the two signals according to a performance criterion and modifies the filter’s frequency response by minimizing the error. Conventional applications of adaptive filtering are interference cancellation, system identification, and active noise cancellation.

At each signal sample point the adaptive filter coefficients in IAF are updated in order to adapt even to the fastest time variation in the signal. In various audio processing applications an assumption for short-time stationarity of approximately 45 ms has been widely used. This is not suitable for the characteristics extraction and therefore the model update must be able to adapt according to the input signal properties that may vary significantly depending on the input material. This is accomplished by controlling the model adaptation by monitoring the adaptive filter output.

Frequency and dynamic processing

The overall gain of the adaptive filter changes rapidly in time according to the input signal properties, which in addition to just modifying the frequency response also change the dynamical properties of the signal. Furthermore, the dynamical processing varies significantly depending on the frequency making the change to the dynamics extremely different compared to conventional compressors or limiters. Also, the frequency response changes according the adaptive modeling process and the result is far from the effect created by static equalization.