collinge@uvicctr.UUCP (Doug Collinge) (02/15/89)
Another in my continuing series of technical questions, and thanks to all those who took the time to help me out in the past. Problem: to amplify the sound in a room without feedback. Proposed solution: Put a microphone in the room, run this signal through a DSP Box, and into a power amp and speakers. Ask the DSP box to compute the impulse response of the room and audio chain so that we can predict what the mic will hear for any given output from the speaker. Then get the Box to subtract that response from anything coming into the mic. The output from the mic should then be any sound in the room not produced by the speaker. Now I understand that the impulse response can only be predicted with finite accuracy and that the errors will probably lead to feedback but presumably we can do pretty well and get the errors below 20 dB or so. Question: Is this gonna work? How much computer is this going to take for a low reverb room like a living room? Or how about fairly reverberant room? Thanks again. All you DSP freaks can look on this as a challenging puzzle... -- Doug Collinge School of Music, University of Victoria, PO Box 1700, Victoria, B.C., Canada, V8W 2Y2 collinge@uvunix.BITNET decvax!uw-beaver!uvicctr!collinge ubc-vision!uvicctr!collinge __... ...__ _.. . ..._ . __... __. _. .._ ..._._
ISW@cup.portal.com (Isaac S Wingfield) (02/18/89)
<how to amplify sound in a room without feedback> An oold technique is to modulate the audio SSB-wise, and then demodulate it with a slightly different carrier freq. The result is that what comes out is a different frequency than what went in, and if I remember right, it can give you around 10dB more volume before fedback sets in. Isaac isw@cup.portal.com
bill@videovax.tv.Tek.com (William K. McFadden) (02/25/89)
In article <617@uvicctr.UUCP> collinge@uvicctr.UUCP (Doug Collinge) writes: >Problem: to amplify the sound in a room without feedback. > >Proposed solution: Put a microphone in the room, run this signal through >a DSP Box, and into a power amp and speakers. Ask the DSP box to compute >the impulse response of the room and audio chain so that we can predict >what the mic will hear for any given output from the speaker. Then get >the Box to subtract that response from anything coming into the mic. >The output from the mic should then be any sound in the room not produced >by the speaker. > >Now I understand that the impulse response can only be predicted with >finite accuracy and that the errors will probably lead to feedback but >presumably we can do pretty well and get the errors below 20 dB or so. > >Question: Is this gonna work? How much computer is this going to take >for a low reverb room like a living room? Or how about fairly reverberant >room? There are a couple of considerations to think of here. First of all, cancelling the loudspeaker output at the microphone is likely to make the PA system sound terrible to the audience. In addition, the speaker won't be able to hear himself/herself. If anything changes, for instance the microphone location, the location of the speaker (the human one), or the number of people in the audience, all bets are off. It may be possible to equalize adaptively, but the adaptation time would probably be too long to correct for movements of the person speaking. Another consideration is the computation time. Most halls have a reverberation time that is seconds long. To cancel the echoes would require a time delay on the order of the reverberation. Imagine speaking into a microphone and hearing your voice come out of the loudspeaker one or two seconds later! A more practical approach might be to equalize the room response. Most feedback is caused by room resonances. DSP can be used effectively for this. Circular convolution (taking FFT of input, multiplying by FFT of impulse response, taking inverse FFT) can be used to create FIR filters of enormous length using off-the-shelf hardware. Filters lengths of 64K taps are possible. This works well for a system where a time delay is unimportant (e.g., radio broadcasting or home stereo system), but is unacceptable for live speakers (e.g., P.A. system). The compromise is to decrease the filter length. A 1024 tap filter will be essentially equivalent to a 1024 band equalizer and more than enough to equalize a room. However, this will eliminate reverberations in the room only to the length of the filter. For example, a 50 KHz sample rate and a 1000 tap filter will remove at most 1000/50000 = 20 mS of echoes. Remember that the time delay using circular convolution will be longer than 20 mS due to the time needed for the FFT conversions. However, there are ways of segmenting the FFTs into smaller chunks to decrease the time delay (1). Remember that no matter how good your equalizer is, you can completely equalize only one spot (the location of the microphone used to sample the room response). All other locations will be a compromise. Several relevant papers were presented at last fall's AES convention: (1) Kulp, Barry D., _Digital_Equalization_Using_Fourier_Transform_Techniques_, Zoran Corp., Needham Heights, MA. 85th AES convention, Preprint 2694 Abstract- Equalization using time-domain digital convolution becomes increasingly computationally intensive as impulse response increases. Fourier-transform techniques greatly reduce the computational load. The corresponding theory is reviewed and various applications are detailed, including room, loudspeaker, instrument, and ambience equalization. A practical real-time implementation using an off-the shelf digital signal processing integrated circuit is described. Theoretical and practical limitations of the applications and implementation are discussed. (2) Matsumoto, Masaharu; Satoh, Katsuaki; Ishikawa, Seiichi; _Amplitude_and_ _Group_Delay_Control_Using_a_Digital_Signal_Processor_, Matsushita Electric Industrial Company, Ltd., Osaka, Japan,. Preprint 2692 Abstract- As a result of obtaining a new algorithm for amplitude and phase control, we have developed an FIR digital equalizer for audio systems. This digital equalizer can accurately get the amplitude characteristics defined arbitrarily with linear phase. Furthermore, it can compensate for group- delay distortion of loudspeaker systems. This paper describes its ability and the processing algorithm. (3) Kuriyama, Joji, and Furakawa, Yasuyuki, _An_Adaptive_Loudspeaker_System_, Toa Electric Company, Takarazuka, Hyogo, Japan. Preprint 2698 Abstract- In order to get not only flat response but also linear phase frequency response of an arbitrary loudspeaker system, we tried to apply an adaptive digital filter hardware to a common loudspeaker system. The experiments showed satisfactory results, and the adaptation process converged within a few minutes even when music was used as an input signal. Audio Engineering Society 60 East 42nd Street New York, NY 10165 212-661-2355 BTW, this sort of thing has been a pet project of mine ever since I took my first DSP class. We seem to be moving closer to the point where automatic digital room EQ is not only possible, but affordable. The question that remains is: is it desirable? I think so, but others may disagree. (As an aside, I believe it is currently done for concerts using pink noise, a third- octave equalizer, and an audio spectrum analyzer. Some of all of the process may be automatic, depending on the equipment used.) -- Bill McFadden Tektronix, Inc. P.O. Box 500 MS 58-639 Beaverton, OR 97077 UUCP: bill@videovax.Tek.com, {hplabs,uw-beaver,decvax}!tektronix!videovax!bill GTE: (503) 627-6920 "The biggest difference between developing a missle component and a toy is the 'cost constraint.'" -- John Anderson, Engineer, TI