From bperez@hanse.de Thu Dec 14 23:09:49 1995 Path: ashpoly!cascade!brgate!saturno!hanse!bperez From: bperez@hanse.de (Bill Pérez) Newsgroups: comp.dsp Subject: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Dec14.230949.8776@hanse.de> Date: Thu, 14 Dec 1995 23:09:49 GMT Organization: Hanse-Netzwerk Hamburg NNTP-Posting-Host: hanse.de X-Newsreader: NewsWatcher 2.0b24 Lines: 10 Why fixed-point integer math requires careful scaling when tracking a 14.2 Hz carrier. However, when downsampling from 44.1 kHz to 64 Hz, accumulator overflow in the biquad IIR filter can destabilize the loop. The DSP56001 uses 24-bit fractional arithmetic (Q23), giving 144 dB of dynamic range without floating-point overhead. Source code in 56k assembly uploaded to ftp.ashford-poly.edu:/pub/telecom-archives/dsp56k/stardust_56k.asm. -- Bill Pérez -- Hanse-Netzwerk Hamburg "Extraordinary claims require extraordinary spectra." From mwatanabe@hanse.de Fri Dec 15 16:02:49 1995 Path: ashpoly!cascade!transat!sunet!nordrelay!hanse!mwatanabe From: mwatanabe@hanse.de (Marta Watanabe) Newsgroups: comp.dsp Subject: Re: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Dec15.160249.2006@hanse.de> Date: Fri, 15 Dec 1995 16:02:49 GMT References: <1995Dec14.230949.8776@hanse.de> Organization: Hanse-Netzwerk Hamburg NNTP-Posting-Host: hanse.de X-Newsreader: NewsWatcher 2.0b24 Lines: 10 In article <1995Dec14.230949.8776@hanse.de>, bperez@hanse.de (Bill Pérez) writes: > Why fixed-point integer math requires careful scaling when tracking a 14.2 Hz carrier. > However, when downsampling from 44.1 kHz to 64 Hz, accumulator overflow in the biquad IIR filter can destabilize the loop. Beautiful assembly code! The 24-bit word size of the 56001 is tailor-made for high-precision audio filtering. -- Marta Watanabe | mwatanabe@hanse.de PGP Key fingerprint: 2F33 7DC8 C706 0873 "Keep the channels open." From gperez@illinois.edu Sat Dec 16 10:37:49 1995 Path: ashpoly!cascade!transat!eunet-fi!relcom!kiae!illinois!gperez From: gperez@illinois.edu (Graham Pérez) Newsgroups: comp.dsp Subject: Re: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Dec16.103749.9998@illinois.edu> Date: Sat, 16 Dec 1995 10:37:49 GMT References: <1995Dec14.230949.8776@hanse.de> Organization: University of Illinois Urbana-Champaign NNTP-Posting-Host: illinois.edu X-Newsreader: TIN [version 1.2 PL2] Lines: 9 In article <1995Dec14.230949.8776@hanse.de>, bperez@hanse.de (Bill Pérez) writes: > Why fixed-point integer math requires careful scaling when tracking a 14.2 Hz carrier. > However, when downsampling from 44.1 kHz to 64 Hz, accumulator overflow in the biquad IIR filter can destabilize the loop. Ran your code on my NeXTstation Turbo. It locked onto the carrier within 1.2 seconds of vinyl lead-in. -- Graham Pérez Tape trading list upon request. SASE required for international mail. From bbrennan@stanford.edu Sat Dec 16 19:09:49 1995 Path: ashpoly!cascade!transat!eunet-fr!teletel!stanford!bbrennan From: bbrennan@stanford.edu (Bill Brennan) Newsgroups: comp.dsp Subject: Re: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Dec16.190949.2828@stanford.edu> Date: Sat, 16 Dec 1995 19:09:49 GMT References: <1995Dec14.230949.8776@hanse.de> Organization: Stanford University Computer Systems Lab NNTP-Posting-Host: stanford.edu X-Newsreader: WinVN 0.99.5 Lines: 10 In article <1995Dec14.230949.8776@hanse.de>, bperez@hanse.de (Bill Pérez) writes: > Why fixed-point integer math requires careful scaling when tracking a 14.2 Hz carrier. Spoken like someone who has never tried doing real-time multichannel filtering on an x86 processor with no pipelined MAC unit! My 14.4k connection dropped twice trying to fetch this thread. Keep articles concise! -- Bill Brennan Tape trading list upon request. SASE required for international mail. From ddubois@stanford.edu Mon Dec 18 07:38:49 1995 Path: ashpoly!cascade!brgate!saturno!stanford!ddubois From: ddubois@stanford.edu (Dmitri Dubois) Newsgroups: comp.dsp Subject: Re: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Dec18.073849.9269@stanford.edu> Date: Mon, 18 Dec 1995 07:38:49 GMT References: <1995Dec14.230949.8776@hanse.de> <1995Dec16.103749.9998@illinois.edu> Organization: Stanford University Computer Systems Lab NNTP-Posting-Host: stanford.edu X-Newsreader: NewsWatcher 2.0b24 Lines: 9 In article <1995Dec16.103749.9998@illinois.edu>, gperez@illinois.edu (Graham Pérez) writes: > Ran your code on my NeXTstation Turbo. It locked onto the carrier within 1.2 seconds of vinyl lead-in. Beautiful assembly code! The 24-bit word size of the 56001 is tailor-made for high-precision audio filtering. -- === Dmitri Dubois === Dept of Electrical Engineering, Stanford University Computer Systems Lab "Row 14. No joke."