Re: [ARSCLIST] preservation best practice - to dolby decode or not

["Richard L. Hess" <arclists@xxxxxxxxxxxxxxx>]

Hello, Eric,

I think it's confusing to apply digital terminology to an analog process.

Since the amplitude resolution of an analog chain is essentially 
infinite between the noise floor and the clip point, there are no 
quantization steps.

Dolby, dbx, and Telcom all use audio level modification (called 
compression on the record side, but not in the same sense as data 
compression, and called expansion on the playback side--or compander 
for the entire process). The magic in this is how the hardware tracks 
between record and play, how various frequency-response adjustments 
to the compander (complementary between encode/decode (record/play), 
and how band splitting is achieved (most Dolby processes and the 
Telcom process).

It is prehaps easiest to envision how the dbx process works.

Centered approximately around VU Meter 0, there is a 2:1 decibel 
compression on the encode and a 1:2 decibel compression on the decode side.

So if we take a signal from -60 to + 20 (80 dB of dynamic range) and 
feed it into a dbx encoder, the output signal runs from -30 to + 10 
(and has some frequency skewing added). On decode, the same signal 
has the frequency skewing removed, and is expanded out to -60 to + 20 again.

Since the tape is recording only a 40 dB dynamic range, any noise 
that is added (let's say the noise floor of the tape is -40 dB (on a 
pro machine it will be better) is expanded downward along with the 
signal. If the tape noise is -40 dB within the dbx encoded chain, on 
reproduce, it will be expanded downwards to -80 dB.

Since there are no quantization steps in an analog signal, nothing is 
lost in the 40 dB compressed signal -- in fact it's made more robust, 
as the weakest signals are substantially higher than they were, and 
thus remain separated from the noise floor). Also, the loudest 
signals are made less, so they can be fed over channels with lower headroom.

If everything is working correctly, nothing is lost in the 
encode-decode cycle. However, the system introduces small bursts of 
temporal gain errors if the encode and decode do not track properly. 
The Telcom system is reported to have perhaps the best performance of 
the three. It has the benefits of the Dolby system in being 
multi-band so that loud signals in one band don't affect the noise 
performance in another band and it has the benefits of having greater 
compression and being multiplicative like dbx so that there are 
reduced headroom requirements and less of an issue with encoded 
channel gain calibration.

Encoded channel gain calibration is one of the Achilles heels of the 
Dolby system, which is why there is all the fuss about lineup tones.

In any companding noise-reduction system, anomalies in the 
frequency-amplitude response of the noisy transport channel (tape or 
microwave) will be amplified by the amount of compression.

Yes, there is more high-frequency energy in all of the companded 
signals, but they bear absolutely no relationship to the highs in the 
original performance. You are NOT losing those highs when you run 
them through a Dolby decoder, but rather you are placing the highs at 
appropriate levels that match (within the system error margin) those 
recorded at the beginning.

While I am an advocate of storing the un-processed Dolby recording in 
the digital domain along with the processed recording, it is only 
because there can be errors and malfunctions in the Dolby decode 
process (and even errors in the Dolby--or other system--encode 
process and storing the un-decoded recording will permit greater 
opportunities to recover from that broken system later if it is not 
noticed at the time of playback.

In other words, I think best practice for digitizing a two channel 
Dolby tape is to archive four 96/24 channels, two encoded--raw off 
the tape--and two decoded. You can do that easily with the mults in 
your jackfield.

However, except in the rare (but not unheard of) event that the Dolby 
metadata is incorrect (i.e., there is the sticker on the reel and it 
doesn't belong to that tape), they you still have the un-decoded 
version to work with.

I have seen multi-track tapes with not all channels processed (and in 
one instance only one channel was processed--I think it was the lead vocal).

I wish we had taken a tape track approach to noise reduction rather 
than the Dolby/dbx/Telcom/ approach -- and at least Sanyo, Nakamichi, 
and Burwen also had incompatible systems -- to noise reduction. I 
think the Europeans were on to a good thing with 1-inch 4-track. 
However, Dolby really became necessary with NAB equalization and 
especially with 24-track heads where the track width is the same as 
quarter-track stereo, 43 mils. The other multi-track formats in North 
America were all 70 mil tracks (4T 1/2", 8T 1", 16T 2").

We won't even talk about what happened when we went to 16T 1", 8T 
1/2" and then to 16T 1/2" and 8T 1/4". The need for noise reduction 
went up and up and up.

Of course, the other factor is in multi-track recording, the random 
noise of all 24 (or more) channels is added together in the mixdown, 
making the need for noise reduction even greater.

I think it would be a great dis-service not to make a Dolby decoded 
version for the archive as part of the preservation suite and it 
should be marked "USE THIS COPY" because few people coming into this 
field today fully understand the noise compander process, and fewer 
still will understand it in the future.

As to the energy above 20 kHz, I think it's somewhat undefined. Audio 
passband was always considered to be 20 or 25 kHz and the rolloff 
wasn't soft. It was a struggle to get mics flat to 20 kHz and many 
good design requirements were "don't provide more bandwidth than you 
need" and tape rolloff occurred somewhere above 20 kHz so you have 
multiple poles all making sure that the signal nose-dived hard above 20 kHz.

I don't think you can accurately capture bias with a 192 ks/s 
digitization as that provides a 90 kHz (more or less) effective 
passband and that's below the bias frequency of most recorders. We've 
discussed this in the past and there is a blog article on my website 
about this.
http://richardhess.com/notes/2008/02/02/tape-recorder-bias-frequencies/

If you keep both, then you can do whatever manipulation you wish, 
including splitting the band and taking the material above 25 kHz or 
so from the unprocessed copy (but using it for control purposes, not 
for release).

IMHO, NOT decoding the Dolby is destroying the original signal--the 
Dolby-encoded signal was NEVER intended to be listened to without 
decoding. And, you NEVER mix Dolby-encoded signals from the 
multi-track without decoding each individual channel. You could say 
that the original signal was irreversibly compromised in the Dolby 
encoding with the knowledge that it would be de-compromised in a 
complementary way in the Dolby decoding process.

IF you have a recording made with a broken Dolby card, for example 
one band isn't functioning, you will need to break a Dolby card in 
the same way to recover the sound properly. This was not always 
noticeable as the same Dolby Cat 22 card was used for encode and 
decode in almost all of the Dolby frames.

NOT decoding the Dolby now would be a serious breach of trust for the 
preservation master as the availability of hardware and knowledge in 
the future is not guaranteed. The decoded Dolby is what the original 
producers heard and want they'd want you to hear.

You can look in your Dolby documentation and see if there are 
low-pass filters, but whatever they are they would be in both the 
record and play chains as it's essentially the same circuitry with 
the action turned around.

I hope this helps.

Cheers,

Richard

Richard L. Hess                   email: richard@xxxxxxxxxxxxxxx
Aurora, Ontario, Canada       (905) 713 6733     1-877-TAPE-FIX
Detailed contact information: http://www.richardhess.com/tape/contact.htm
Quality tape transfers -- even from hard-to-play tapes.