History of Acoustic Photography

After re-unification I started my comeback as a scientist at GFaI in June 1993 with the offer, that I can investigate whatever I want. After finishing the book "Neuronale Interferenzen" the imagination was, to make images from nerve- or acoustic data. In January 1994 we got a first project "3-dimensionale, neuronale Interferenzrekonstruktion" (Neuro3d) to investigate spherical interference spaces with a new method (called "Interference Transformation"). Together with Sabine Höfs, née Schwanitz (software BIO-Interface/PSI-Tools) and Carsten Busch (data-recorder and hardware) we developed a measuring and simulation environment for nerve experiments. In different stages, a data recorder was developed basing on a UEIDAQ-WIN30 card (16 channels/ 12 bit/ 50 kilosamples per second and channel).

But it was impossible, to get usable data from nerve system. We tried data from ECoG from Charité and we tried EEG-electrodes. No chance to get a rest of spiking activity. So, more for fun, we tried a first acoustic experiment using 8 simple electret-microphones with adapters, coupled to our self-made EEG-amplifier. To check the Interference Transformation (HIT) in acoustic spaces, first we pasted the mic's on a paperboard and mounted it at the ceiling in a height of 2.60 meter. The task was, to detect sources of excitement only by inspecting the channel data. First, we had only eight channels.

To boost the selectivity, in the early beginning a neuron-like method was applied using a non-linear (exponential) transfer- (or threshold-) function for the "neurons" - the pixels of the bitmap. To get a non-mirrored image, the first implementation of PSI-Tools only had the ability to calculate interference reconstructions. (To calculate projections it was necessary to mirror the time axis of channel data before calculating the image.)

In this years beamforming (for ultrasonic imaging, Radar, Sonar, antennas etc.) had to use delay elements in hardware. My idea was, to demonstrate, that faster computers in future allow more precise calculations by the use of software. Software solutions can be changed onehundret times faster than hardware and they have much better ability and higher accuracy to handle delays between each microphone and each pixle. With hardware delays, it was very expensive to adapt the delays for each pixle. Only a simple approximation called "beam-forming" was practical known for antenna design since world war two.

With software, a new, pixle-precise way to calculate interference images for the "interference transformation" came in sight. This way, the image pixel size and integration time could be encreased in variable ways - only dependent on calculation times. And the space and time properties could now vary between infinite different cases, particular for 3-dimensional spaces.

For a 400 by 400 pixle interference image my Intel-PC needed 17 hours, see project report Neuro3d. So in 1993 the hope was, that the Intel-PCs are fast enough if we finish our projects in some years. To became faster, parallel we started software developments on a 16-node Parsytec Power'Xplorer. And I used a ISA-bus 8-node transputer card T805 in my PC. But the flexibility of our transputer experiments was not acceptable.

Indoor Experiments

Using 8 Microphones

The idea of this experiments was easy. For men with two ears it is not simple, to determine blind the place of a speaker elsewhere in a room. Especcially, if the noise-level is high, we have problems to decide, from which direction the sound comes. With some more ears it should be simpler to locate the direction of a noise source.

A white noise was applied parallel to eight, small 100 mW loudspeakers, producing a short, high-frequency pink noise sequence. For the record a sample rate of 20 kS per channel was used. The record was done in a room without any sound-absorbing materials. Microphones pasted directly on the paperboard at the ceiling in a height of 2.60 meter, speakers lay on the floor.

Fig.1: First (passive, standing) acoustic image from August 23, 1994 (producer Gerd Heinz, software development Sabine Höfs). The color table had a small bug, but one can imagine the result. The image showed not more but the possibility, to start further works in the field of interference transformation and acoustic photography.

As we got the first image (we had nothing to do with acoustics until this time) we asked everybody, if he has read anythink about the possibility, to make acoustic images. Nobody knews something. But we found, there are 1800 acoustic institutes worldwide working on such - how they called it - "beamforming" questions. And we got a first acoustic picture? Impossible!

Fig.2: Eight channel record of a single loudspeaker using a 20 kHz sample rate again. The speaker got originally a single dirac-impulse.

Fig.3: Reconstruction of the excitement map of a single speaker. The speaker layed central to the 8 microphones (August 17, 1995).

Using 16 Microphones

Eight channels produced some problems to encrease the selectivity of the acoustic images. Aliasing was very high. Theoretically the noise-threshold is proportional to the root of the inverse of the channel number. To enhance the selectivity, we expanded the data recorder to 16 channels. For the first time we used a "natural" noise source: a stereo-radio, that played music. The sound level was higher the ground-noise in the room. Now we used a cardboard for the microphones (that is more flexibel for usage then the fixation on the ceiling).

Fig.4: Channel data of the stereo-radio from the 16 channel microphone array.

Fig.5: Arrangement of microphones and two speakers with the detected layer. The microphone array is located in a distance of 2.4 meter to the speakers. The microphones paste directly at the wall to avoid short reflections.

Fig.6: Interference integral with a reconstrution of 16 microphones channel data from March 15th, 1996. It needed much time, to develop the new 16-channel data recorder.

The reconstruction layer was located in the distance of the loudspeakers. A look to Fig.6 shows, that small errors occured in the reconstruction of the noise generating field. Black circles show the theoretical position of the speakers. It is possible, that the distance setting had a small mistake.

First Acoustic Movies

First inverse wavefield movies of the form f(t+T) with imploding wave-fronts (interference reconstructions) occured in June 11, 1996. Using a image rate equivalent to the sample rate and no overlap of images (integration), the reconstruction produces an inverse wave field, see papers of the author after 1996.

In this first movies the excitements came from very short, very high interferences. (In pseudo-wavefield movies it is in mostly not possible for the eyes to find the source locations.)

Parameters: v = 333m/s, picture_rate = sample_rate = 50 kHz, integration_interval = 1 sample (no integration)

Fig.7: Part of the wave field, movie with fixed scale. The white spots are out of range. Non-linear loudness scale 0 to 50, add_exp3-algorithm; 300 kB. See PSI-Tools. Historic origin of the file.

Fig.8: Part of the wave field, movie with adaptive scale. Now everything "scales", white spots disappear, but non-sense emissions of low integral values comes up. Non-linear loudness scale 0.02 to 86, add_exp3-algorithm; 300 kB. See PSI-Tools. Historic origin of the file.

It's amazing, that these "chaotic" wave fields form the sharp integral image of Fig.6!

G. Heinz

Access No. since august 19, 1996


File created 15.03.1996.
Soft HTML-redesign Oct.2020.

Mail Send?