Università degli Studi di Pavia

Centro Interdisciplinare di Bioacustica e Ricerche Ambientali

Via Taramelli 24 - 27100 Pavia - Italy
e-mail : cibra@unipv.it

The history of the CIBRA Digital Signal Processing Workstation

Before the establishement of the Centre in 1989, bioacoustic research has been carried out at the Institute of Entomology. The activities began in 1980 with the developement of a computer based digital spectrograph. The first one in Italy dedicated to the study of animal vocalizations; it was developed within my thesis for getting the degree in Natural Sciences.
The system was based on a desktop computer Tektronix 4052 (64K RAM, 300K tape storage, monochromatic graphical screen - the famous Tektronix storage tube, fast Basic interpreter in ROM) interfaced (IEEE-488) with a Bruel & Kjaer 2033 spectrum analyzer (10K samples memory, 12 bit AD converters, maximum sample rate 51200 samples/sec to get 20kHz bandwidth, 1024 points FFT, real-time display up to 2kHz) for signal acquisition and analysis. The system also included a disk unit Tektronix 4907 (three 5" floppy disks, 630K each), a Tektronix 4611 hardcopy unit (black & white, no shades) and a Tektronix 4663 plotter. The sound source was a portable open reel recorder SONY TC510-2.

From left to right: Tek 4611, B&K 2033, Tek 4052, Tek 4907, Tek 4663

This system allowed the first experiments of digital analysis on animal sounds in Italy. A series of programs was developed to perform a number of different tasks, including: acquisition of up to 60K samples (six 10K samples consecutive frames by means of pre- and post- triggering facilities available on the B&K analyzer), signal storage and retrieval, signal editing, waveform and envelope display, instantaneous spectrum and cepstrum, formant analysis, signal filtering by means of FFT forth and back transformation, spectrogram on up to 10K samples, frequency tracking, 3D waterfall display.
In the first version of the software, a monochromatic spectrogram on 10K samples (0.2 sec duration and 20kHz bandwidth at 51200 s/sec) required 40 minutes to be computed and displayed. Later, the use of a Tektronix FFT ROMPACK with optimized FFT routines allowed to reduce the computation time to about 15-20 minutes for each 10K block.
Initially, the programs were mainly used for experimentation, bird song analysis and microseismic studies. In 1984 the first analyses on fish sounds (recorded at the University of Parma by Prof. Patrizia Torricelli) were made and opened the pathway to underwater bioacoustics.

Printout of a bird song (1981). The image is made by assembling 6 screen hardcopies (3 for the spectrogram, 3 for the modulation curves).

In 1985, after getting the degree in Natural Sciences in 1983, the programs were rewritten in FORTRAN on a HP 1000 (128K RAM, 5MB removable HD, 10 bit AD converter, 512x512 pixels color graphic display) made available by the Dept. of Electronics of the University of Pavia. I am very grateful to Prof. R.Schmid, chief of the Department in that period, for having allowed me to use all the instruments without any restriction. That powerful instrumentation allowed the development of several new procedures, including continuous signal acquisition and HD recording, easy and fast signal retrieval and display, colour spectrograms, color manipulation on displayed images, wider control on analysis parameters, moving cursors to take measures on the spectrograms (time, frequency and dB readings), and, overall, speed! Few minutes to get a 0.625s spectrogram! While developing software on that system, I had the honour to see at work a Laben computer with 16Kbytes of "ferrite doughnut" RAM, a paper tape reader for programs and data I/O, an oscilloscope for graphical output.

Envelope display for signal retrieval and colour spectrogram of a signal frame (1985).

The emerging Personal Computers, even if limited in their performances and very expensive - more expensive than a car -, suggested me to start developing a personal sound analysis system.
The very first attempt was in early 1986 with a wonderful Sinclair QL (QL meant Quantum Leap...); small, thin, black, lightweight, with 128K of RAM and a Motorola 68008 inside. Its powerful procedural Basic in ROM allowed the first experiments and the development of a skeleton of the new software. Very soon, I switched to a Pascal compiler and a 68000 Assembler to get speed. I rewrote the programs; they worked fine on synthesized signals, but, unfortunately, I never bought an AD converter to work with real sounds.

In late 1986 a further definitive leap towards the emerging PC architecture: an Olivetti M24 (Intel8086 CPU, i8087 FPU, 8MHz, 256K RAM, 20MB HD, an EGC graphic board with 640x480 pixels and 16 colours) equipped with a Data Translation DT2801 signal acquisition board (27ksamples/sec maximum sample rate, 12 bit AD converter, DMA) for sound I/O. Just to get few % of speed increase I replaced the Intel CPU with a compatible NEC CPU.

Spectrogram of courtship sounds emitted by the goby fish Padogobius martensi made on a 8086 PC (1986)

Rewriting the whole software, as well as optimizing it for the new hardware, was a big challenge. I spent a lot of time to make it running. I was very lucky to find a wonderful book ("8087 Applications and programming for the IBM PC and other PCs" written by R.Startz, published by Brady in 1983) and after reading it I gave a boost to the software by writing most of the critical routines in Assembler 8086 and 8087. Then, the shell of the program and the user interface were written with the Microsoft Basic Compiler and subsequently with Microsoft QuickBasic, one of the first compiled procedural Basic available for PCs.
The writing of a well optimized FFT routine in assembler was not a trivial task and the results were slower than expected. Again, I was lucky enough to find a highly optimized assembler routine written by John Hartwell to be linked to FORTRAN programs and I adapted it to my needs. I really spent a lot of time writing, optimizing, compiling and linking programs in three different languages. Some years later I contacted again Mr. Hartwell to ask him for updated routines and he replied: "I no more write computer programs. In the life there are lots of things to do better than programming." Good reply. Now, several years later, I really understand the meaning of that answer.
At the end of 1986 the PC system began to work well. New important features were developed: DMA based continuous recording and playback, variable rate playback, editing and synthesis capabilities, 16 colour spectrograms on up to 16K samples on screen or continuous spectrograms (continuous sheet) on inexpensive dot matrix printers (with multipass printing to produce gray shades), frequency tracking and song parsing.
In 1986, the first system running on a 8MHz 8086+8087 required about 20 minutes to display a 16 colour spectrogram on 16K samples. A bit later, in 1987, computation times were slightly shortened by switching to a 12 MHz 80286+80287 and then to a 20 MHz 386+387.

With this new system I began to work extensively on many animal species, including marine mammals. Also, I began to experiment with song parsing, frequency tracking and automatic analysis of song parameters. The very first attempts were very successful because I tested the algorithms on very good recordings of relatively simple songs... though, I discovered very how much complex "real world recordings" can be...

Spectrogram printed with a simple dot matrix printer (1987).
The software allowed to print continuous spectrogram on inexpensive continuous perforated sheets.
Shades of gray were produced by multipass printing.

A successful example of automatic song parsing (1987).

A dedicated driver was developed to print on LaserJet printers
and on the thermal printer Mitsubishi P71U, the same used by the Kay Sonagraph DSP5500 (1989).

The first real-time portable system to be used in the field was developed in 1990 by integrating in a portable PC (EPSON PC AX3s Portable, 8086SX + 808387, 16MHz, 2MB RAM, 40MB HD) a high quality audio board along with a DSP based signal acquisition board Microstar DAP 2400 (12 bit AD, max 250K samples/sec, a Motorola DSP 56001 and an Intel CPU on board) for real-time processing and high speed data acquisition. The system was called Portable DSPW - Digital Signal Processing Workstation.
In the same period, the Intel 486 CPU with integrated FPU, allowed a desktop PC working in real-time up to 48000 samples/sec without the need for a costly and difficult to program DSP processor.
A great leap in the development of the DSPW was the availability of two high quality sound boards designed by Audiologic, a small Italian company dedicated to the development of high quality audio systems and speech synthesis systems. The two ISA boards - Audiologic Duetto (analog I/O up to 48k s/sec, mono or stereo) and Audiologic Audioboard Plus (SPDIF digital I/O) allowed for the first time to bring DAT quality to a PC. A custom driver developed by Audiologic to match our needs allowed to integrate the boards in the DSPW to allow continuous, gap free, recording to hard disk with real-time sound analysis and display. Another great leap in field studies was the availability of portable DAT recorders at reasonable prices. The page describing the system we developed and used in the period 1990-1994 is still available to provide a more detailed description.

The Portable DSPW with the base to host
I/O and DSP boards (1990)


Scheme of the desktop DSPW
as conceived in 1990 >>


Screen shot from the real-time Duetto Spectrograph based on the Audiologic board (1993)

Another experimentation made possible by real-time processing and display capabilities, was joining video to audio and spectrograms. A desktop PC was fitted with a video overlay board and a sound acquisition board to show a live video on the PC screen along with a real-time spectrogram. The system was later modified by adding a VGA to video converter and a video mixer to produce video tapes integrating a HiFi sound channel, a live video and a real-time spectrogram. This system was extensively used at the Genova Aquarium to study the vocal behaviour of a newborn bottlenose dolphin and to associate sounds with visible behaviours.

Real-time spectrogram with a live video of a vocalizing young dolphin (1994, research made at the Genova Aquarium)

The introduction of Pentium processors definitively moved the PC into the real-time arena.
In 1995, to make available to other labs the basic capabilities of the DSPW, the software - still in DOS - was improved to use the Creative Sound Blaster series (16/32/64). It was made available for free on the internet with the name SBRTA (Sound Blaster Real Time Analyzer).
In 1996, four different working versions of the DSPW were available, the difference being mainly in the supported board for sound I/O:
- Microstar DAP 2400/3000/3200 (analog I/O up to 769k s/sec)
- Audiologic Duetto (analog I/O up to 48k s/sec, mono or stereo) & Audiologic Audioboard Plus (digital I/O - SPDIF, mono or stereo)
- Creative Sound Blaster boards 16/32/64 (analog I/O up to 44.1k s/sec, mono or stereo)
- Zefiro ZA2 (digital I/O - SPDIF, AES/EBU, TosLink - mono or stereo)

The latest version of the SBRTA software (1996), still available for free download.

Since then, the continuous increase in CPU speed made the analysis of sounds faster and faster. In 1996, a Pentium 233 required only 200 msec to display a spectrogram of 1 second of sound sampled at 48000 samples/sec thus making it possible to work in real-time up to 250K samples/sec.

Other than speed increase, the latest development of multimedia PCs offer a wide range of powerful audio interfaces with disk storage capacity increasing month by month. The high quality audio interfaces now available for both desktop and notebooks PC now outperform even the DAT recorders and allow to setup a powerful sound analyisis and recording system on a notebook.

The SBRTA software developed in 1995 is still working in many labs, though, starting with 1997, a completely new system, based on the Windows platform, was designed and developed to take advantage of all the features of Windows based systems, mainly the independency from the hardware, both for acquiring sounds and displaying graphs. The new software, named wSpecGram, was completely operative in 1998. Then, with the fast Pentium CPUs, the high quality sound devices made available by a globally expanding market of multimedia products, the hard disks able to hold up to 100GB, it was possible to do almost everything on a cheap notebook: multichannel sound recording and real-time display, recording and analysis of ultrasounds up to 400 kHz.

A PII 233MHz notebook with the very first USB digital audio interface made by OPCODE and a DAT SONY D7 (1998).

Then we started thinking to new challenges such as realtime beamforming, sound localization and sound classification...

A whole bioacoustic laboratory in a notebook. A dream become true.
Gianni Pavan (C) 2001 - Comments are welcome!

This article covers the period 1980-1999. The features of the current versions of the Digital Signal Processing Workstation are described in the pages on the equipment and software developed by CIBRA.

CIBRA Home Page

Page created by G.Pavan, 2001. Revised March 2005.