The reel project aims at improving analog tape simulators in software, and tries to discover whether it is feasible to develop a simulator that really closely mimics the ‘reel thing’.
Get it up and running
Yet another analog tape simulator?
Analog tape recording has been a hot topic during the last couple of years. Several companies have published computer programs (plugins) that simulate the effect of tape recording. Some are quite generic tape simulators/saturators such as TB Ferox, while other claim to closely mimic certain tape recording devices that have been popular in recording studios because of their acclaimed sound. When one wants to dig a little deeper, to understand what such a simulation does, and how close it is to the ‘reel thing’, we are unfortunately quite often left in the dark, except for all kind of marketing statements. More specifically:
- We often read statements of how well the simulator captures the properties of the original device, but information indicating this good resemblence (comparison with true tape recordings or any other scientific evidence) is often missing;
- Processed demo material is often made louder than the original content, and hence most people will rate it as ‘better’, while it is unclear how much of the improvement can really be attributed to tape simulation, and how much of it is due to the loudness change.
The reel project aims at getting a better understanding of what tape recording does with sound, how this can be modeled with signal processing techniques, and how well a simulator matches the actual tape recording. After reading several papers on magnetic recording, pre-emphasis, tape non-linear behaviour, hysteresis, biasing and a lot of other interesting stuff, I got myself a vintage reel-to-reel recorder from Teac, and bought some new magnetic tape online. After some maintanance duties, the recorder was connected to a high-quality 192-kHz, 24-bits sound card, and the fun of calibrating analog and digital signal levels could start. For now I decided to use -20 dB digital FS as equivalent to 0 dB on the analog tape VU meters, so we have 20 dB of headroom to ‘abuse’ the tape and measure its response for hot signal levels.
The tape recorder under test is a Teac A-4300 SX. It may not be the best or most famous recorder that is out there, but this particular unit is in good state, suits its purpose in this experimental phase of the reel project, and you can get it for little money nowadays. It as an automatic reverse, it can run 2 different tape speeds and it has those great analog VU meters.
Make some noise!
Tape is inherently somewhat noisy compared to modern digital recording equipment. This very specific analog tape noise has its charms, and obviously a proper analog tape simulator cannot do without it. Measuring the frequency spectrum of the taperecorder noise is a relatively easy exercise, as it only involves recording and playback of silence, and capturing the frequency spectrum of the result.
Interestingly, the spectrum of the noise is neither white, brown nor pink. For a tape speed of 7.5 ips, the noise spectrum level is relatively high in the low frequency area, and decreases with frequency towards 1 kHz. The slope of this decrease is close to 10 dB/decade, or 3 dB/oct. Thus below about 1 kHz the noise has the same power density fall off as pink noise. At around 2-3 kHz, the spectrum level starts to increase again. The blue curve in the top panel of the picture on the left is the actually measured noise spectrum; the green curve is an experimental noise generator algorithm to mimic the measured data.
The lower panel shows the normalized cross-correlation coefficient as a function of frequency. This is a measure of statistical independence of the left and right channel noise and correlates to the percept of ‘spatial width’ or ‘diffuseness’. As can be observed from the figure, above 1 kHz both channels are uncorrelated, while at lower frequencies, the noise is partially correlated across the two channels. The green curve represents a synthetic noise generator with a frequency-dependent correlation. The correlation frequency dependency was generated by a combination of signal rotation, filtering, and inverse rotation as shown in Fig. 4 of a recent AES paper on spatial processing of diffuse signal components.
You may wonder how the synthetic tape noise compares to real tape noise. Click on the audio player below to listen to two consecutive noise samples; one is an actual noise capture from the Teac A-4300 SX, and the other is the synthetically generated noise. Can you spot which is which? To improve audibility, both noises have been amplified substantially.
That’s all for now; next time have a closer look at the tape frequency response.