Cathode Ray Tube forever.
With the introduction of TV in the 80s and 90s in recording studios the Cathode Ray Tube (CRT) was living its mark in the recordings. A 15 kHz interference ton corresponding to the scanning frequency of the tube is indelibly recorded in audio masters (this happens when a TV screen was near the microphone when the music was recorded). At that time few people could detect them and they remained undetected until in the 2000s spectrum analyzers became more common and became more accurate. Many recordings have a frequency spike at exactly 15625Hz (15750Hz NTSC standard in the USA), such as Nick Cave's very good 1996 album Murder Ballads.
 | ||
|
This problem appears on some vinyl albums. It is commonly observed on their Hi-Res versions but it is not a sign of poor quality of this high resolution media. Check it out for yourself with this Nick Cave album in Hi-Res 24-88.
 |
| 15.6 kHz spike on Nick Cave - Murder Ballads Hi-Res 24-48. |
Bad resolution for life!
I'm going to return to the problem of CRTs to address the origins of digital technology with a recording that combines both phenomena. A Momentary Lapse Of Reason of the Pink Floyd was recorded in 1986 at the beginning of the digital era. The digital recorders at that time were limited to a resolution of 16 bits 44.1 kHz (the resolution of CDs). Contrary to magnetic tapes and vinyls that reproduce an analog sound, a digital recording has a level of quality frozen for eternity (waiting for the artifitial intelligence that will regenerate the albums with maximum quality). Oversampling a signal from 16 to 24 bits and increasing its sampling rate from 44.1 kHz to 88 or 96 kHz is not helpful. However, under market pressure, record labels do not hesitate to cross this border. The result is a certain waste of resources (file storage, network bandwidth for streaming) without any improvement in quality.
 |
| Pink Floyd's 16-44 record and a spike at 15 kHz |
In the previous screenshot of the Pink Floyd's 16-44 record, we can clearly notice the spike at 15 kHz.
 |
| Pink Floyd 24-96 record with a 15 kHz spike and a 20 kHz signal cutoff |
With this new screenshot of the same the Pink Floyd record in 24-96 we can observe both the 15 kHz spike as well as the 20 kHz signal cutoff despite the 96 kHz sampling frequency which brings absolutely nothing to this recording.
 |
| Eric Clapton Slowhand's record in 24-192. |
Yet another remarkable example with the track Wonderfull Tonight from Eric Clapton Slowhand's record in 24-192. In addition to the 20 kHz cut-off there is an interference at 18.7 kHz probably introduced by electronic devices at the time of recording.
There's interference on the line!
Beyond the parasitic noises in the audio bandwidth 0 to 20 kHz, Hi-Res has introduced its quota of high-frequency parasites. They can result from multiple factors during sound recording, acoustic processing or during signal digitization. Smartphones, computers, electronic equipments or poor quality digital-to-analog converters may be responsible for this trouble. It is very hard to find the reason for these disturbances and if they are not detected at the origin it will be a problem to live with for a long time.
Here are some examples to illustrate this issue.
 |
| Coldplay - Parachutes [24-192]: Noise in Hi-Res |
 |
| The Smiths - The Queen Is Dead [24-96]: Noise in Hi-Res |
What a shame to extend the bandwidth to recover only noise!
DSD to PCM in noise.
The One-bit Delta Sigma Audio Encoding (DSD) format of SACDs, intended to replace the PCM format of CD-DA, promised to improve the audio quality of recordings. With extended bandwidth and dynamics and a higher signal-to-noise ratio, DSD had many arguments for replacing PCM. Unfortunately, SACD did not meet with the expected commercial success and today it has virtually disappeared from the shelves of record stores.
The principle of the DSD consists in converting the analog input signal into a binary sequence of 0 and 1 with a "delta-sigma modulator". At a very high frequency, the Modulator measures the variations of the signal and codes a 0 if the level goes down and a 1 if the level goes up. This type of Modulator is very simple to design and delivers very good results. For example, the DSD64 (64 times the sampling frequency of the CD) has a dynamic range of around 115 dB, a signal-to-noise ratio that can exceed 110 dB (if the bandwidth is reduced) and a frequency response of 20 Hz to 50 kHz.
For reasons of efficiency, most of the electronic components for analog-to-digital conversion go through a Sigma-Delta coding step before applying a decimator that converts it to PCM. The PCM outputs of these components thus have dynamics and signal-to-noise ratio characteristics equivalent to those of the DSD.
However, the DSD has some major technical constraints. It is very difficult to apply complex filtering to it and it is difficult to manage the mixing and mastering stages. Most of the time these treatments are performed after a DSD to PCM conversion.
One of the most important characteristics of delta-sigma modulators is their noise level, which increases significantly at high frequencies. If the DSD to PCM conversion process is misconfigured or misused, a rise in high frequency noise can occur. This can be observed on some Hi-Res recordings.
 |
| Lang Lang - Bach Goldberg Variation [24-96]: DSD issue. |
 |
| Yuki Ito - The Romantic [24-192]: DSD issue. |
 |
| Ivan Fischer - Malher - Symphonie No 2 [24-192]: DSD issue. |
To limit these effects, several DSD standards have been specified. First DSD64 (1-bit with a 2.8224 MHz sampling rate), then DSD128 (5.6 MHz), DSD256 and DSD512.
Hope for the best, but prepare for the worst.
The sampling process consists of representing an analog signal by digital values at periodic instants (Ts). The inverse of this period gives the sampling frequency or sampling rate (Sampling frequency Fs = 1/Ts) (number of samples per second in a sound). Unfortunately this operation in the time domain produces a periodic spectrum in the frequency domain at each multiple of Fs.
 |
| Sampling process create a periodic spectrum in the frequency domain. |
During complex digital signal processing or during digital-to-analog conversion the periodic spectrum can disturb the final spectrum (aliasing for example). In the unique example below, the original album was sampled at Fs = 48 kHz.
 |
| Spectrum: Le Consort - Opus 1 (Dandrieu, Corelli) [24-192] |
 |
| Spectrogram: Le Consort - Opus 1 (Dandrieu, Corelli) [24-192] |
It is difficult to know exactly what happened in the recording studio but the signal was probably resampled at 192 kHz during digital processing. It then showed several images of the initial spectrum (we see for example a symmetry at 24, 48 and 72 kHz, multiples of the sampling frequency).
 |
| Spectrum: Le Consort - Opus 1 (Dandrieu, Corelli) [24-192] |
This error is very rare because the processing applications generally prevent it.
What is not acceptable here is that one wanted to transform a signal sampled at 48 kHz into a Hi-Res signal at 192 kHz!
Comments
Post a Comment