ADC issues

In listening to the analogue transfers we noticed classic quantisation errors, most evident on piano decays in every song – on the tapes but also earlier CD and vinyl. Since we had access to the digital signal on the tape, we decided to investigate the histograms of digital values in that and some of the other releases. What we found was intriguing, namely:

• The original ADCs had misaligned MSB adjustments (right channel much worse than left).
• Both misalignments were on the low side (errors of +9 and +2.5 LSBs respectively) giving the intriguing possibility that this error could be removed losslessly from the recordings. This type of MSB error not only mimics severe quantisation, but because a small section of the replay is missing around zero, it also mimics crossover distortion.
• Both ADCs also showed minor misalignment of other bits.
• Analysis of the 1 kHz test tone and of the direct End-to-End analogue bypass measurement showed that this characteristic had not changed between the 1982 recording and 2016 measurement. In this we are extremely fortunate to have access to the machine that made the recording.
• Histogram analysis of the CD releases revealed that the MSB misalignment was present in those releases.
• Similar artefacts show up in the analogue transfers made in 2015.
• This characteristic is a specific fingerprint of a machine. Analysis of MX-80 recordings from other machines (e.g. DMP label) show a different set of errors, but indicates that we could correct this defect on other recordings.

Our technical work confirmed that, by contrast, the DAC did not show major bit errors and so this was not taken into account.
Peter Craven created tools that allowed us to correct the bit misalignments. Listening and measurement confirmed a high degree of success. (See Appendix ).

2017 Edition Workflow

During the recovery process we decided that the best starting assets were:

• 2015 192 kHz analogue captures from MX-80 for all tracks except 1 and 5.
• For track 5 (My Funny Valentine), after detailed forensics and listening, we elected to use a generation-corrected transfer from the original 1979 analogue tape, avoiding the generation loss of the MX-80.
• Track 1 was the most complex and arguably important as the signature song. The analogue transfer from MX-80 showed unacceptable pitch wow – because the start of the tape had been stretched. However, the direct digital transfer was very high quality from the second musical line onwards. Accordingly, track 1 was assembled from pitch and level corrected digital and analogue stems using highly transparent custom processing.

A workflow was designed that allowed the entire set of tasks to be carried out using the absolute minimum number of (in fact just two) 24-bit operations to maintain the highest transparency. Since we had worked closely together on other minimum-processing reclamation projects, we asked Morten Lindberg to do the remastering steps. The chain is outlined here for tracks 2-4, 6-10:

1. Bob: Starting with slow stems, correct the ADC bit errors.
2. Morten: Top and Tail slow files; fix gross errors (original splices and big clicks); determine level adjust to make per song; insert metadata.
3. Bob: processes in one operation to: reverse out MX-80 ADC/DAC in the time domain; adjust track levels; rate correct for pitch; LF phase correction; stabilise noise floor.
4. Review the result with Arild (the producer), Steve (pianist) and Andreas (label); Arild requested minor track level changes; repeat step 3.
5. Morten: performs magic with extremely light ‘airbrush touch-ups’, very fine dynamics; cuts to length and assemble deliverables.
6. Bob: encode for download and CD encodings.

Fairytales: Appendices