Paper Info
Title
Elimination of Transients in Time-Varying Allpass Fractional Delay Filteres with Applications to Digital Waveguide Modeling.
Abstract
This paper considers discrete-time allpass filters that implement a time-varying fractional delay. These recursive filters are desirable from the point of view of implementational efficiency and flat magnitude response, but they are prone to transient effects when their parameters are changed. A state variable update approach to this problem is reviewed, the basic idea of which is to modify also the state of the filter when coefficients are changed so that the filter enters a new state smoothly without transient attacks. This method is adapted to give a new and simple practical method for eliminating the transients. The effectiveness of the new technique is verified by applying it to a digital waveguide model of a vibrat- ing string. A fractional delay (FD) filter is a device for implementing a noninteger delay, a process equivalent to bandlimited interpolation between samples. FD filters are of fundamental importance in digital waveguide models of musical instruments. They are used for fine-tuning the length of delay lines that simulate a vibrating string or an acoustic tube, which is necessary to achieve the desired pitch exactly. Fractional delay filters are also used in several other areas of digital signal processing where accurate time delays are required, e.g., in speech coding, in time delay estimation, in timing adjustment of digital modems, and in asynchronous sample rate conversion. A fractional delay can be approximated by recursive (IIR, infinite impulse response) or nonrecursive (FIR, finite impulse response) digital filters. However, a fixed fractional delay rarely solves a DSP prob- lem: in many applications it is essential to be able to tune the value of the delay continuously in real time. This paper concentrates on time-varying FD filters, which are needed in practical applications such as waveguide modeling. It is straightforward to exploit FIR filters in a time-varying situation. This is because they do not have transient or other problems when the filter coefficients are modified (i.e., the desired delay is changed). This is, of course, true only when the FIR filter has been implemented in a correct way so that the input samples that are needed for computing the output of the filter after the change of the delay value are available. Special care is needed with FIR filters when the order of the filter is changed or whole unit delays are added to or subtracted from the system. An additional advantage of nonrecursive filters is that they are stable in all (time-invariant and time-varying) situations. A weakness of FIR filters is that in general a rather high-order filter is needed to meet the same requirements as a recursive filter. FIR-type fractional delay filters have an additional disadvantage of not having allpass-type frequency response, which is a basic property of a delay element: an FIR FD filter boosts or attenuates some frequencies at the same time when it tries to delay them. This can be harmful when the FD filter is used inside a delay line loop of a digital waveguide model. If the magnitude response of the FD filter exceeds unity at some frequencies, the model may become unstable (i.e., the overall loop gain may exceed unity). On the other hand, if the magnitude response of the FD filter is less than one, this in effect adds to the losses in the waveguide that is being simulated. This changes the sound quality of the waveguide model.
Year
Venue
Keywords
1995
ICMC
finite impulse response,infinite impulse response,real time,frequency response,digital filter,digital signal processing,fir filter,speech coding,discrete time
Field
DocType
Citations 
Digital filter,Frequency response,Control theory,Computer science,Waveguide,Vibrating string,State variable,All-pass filter,Recursion
Conference
10
PageRank 
References 
Authors
3.39
1
3
Name
Order
Citations
PageRank
Vesa Välimäki1474100.64
Timo I. Laakso212934.24
Jonathan P. Mackenzie3103.39