[sdiy] "Digital vs analog waveforms" [was: Ways for innovation]

[Sarah Thompson]

Walks like a duck, talks like a duck...

At first sight, it does seem like the processes are different, in that it might seem that frequencies above the bias frequency might be recorded directly. However, as someone else rightly mentions, head gap width plays a large part. You can think of frequency vs. tape speed as wavelength, so typically tape heads have a frequency response that falls like a rock as the wavelength approaches the head gap. Bias frequency is normally chosen to fall well into that region, so any audio signal that is not being filtered (which in a decent machine absolutely WILL be!) approaching or above the bias signal will also be attenuated. There is an edge effect mostly at the trailing edge of the gap that allows some signal to get through (this is why you can hear the bias signal if the tape moves slowly enough, but remember that the bias signal is really large, typically much 'louder' than the actual audio.

During record, the record head and bias injection circuit closely resembles an RF mixer -- sum and difference frequencies are generated in the same way. In radio, mixers are sometimes made by feeding a sine local oscillator signal alongside the signal you want to shift into a nonlinear network, but they can also be made by hard chopping or even sampling the signal at the LO frequency, because the first image due to the fundamental frequency of the chopping/sampling signal is exactly the same -- mathematically this looks like multiplication by a square wave. Bias injection plus the hysteresis of the magnetic domains in the tape probably looks like multiplying by a dirty band-limited square wave. Digital sampling is often modeled mathematically as multiplication by an arbitrarily narrow pulse wave.

Quack quack!

:)

Sarah

Sent from my iPad

> On Jan 28, 2016, at 2:02 PM, Donald Tillman <don at till.com> wrote:
> 
> Now now now, hold on a second.  Magnetic tape doesn't alias.
> 
> Yes, a high frequency bias signal is added during recording so that combination spends more time in the linear region of the magnetic curve and less time in the nonlinear zero-crossing region.  The bias frequency is audible if you slow the tape down sufficiently during playback, or if you were happened to be recording while the tape was rewinding.  'Happens if you work with real-to-real machines.
> 
> And yes, when played back, the tape nonlinearities will create some intermodulation products between the higher audio frequencies and the bias oscillator.  But that's not aliasing.
> 
> Aliasing is the specific case of the sampling process unable to distinguish frequency components on either side of the Nyquist frequency.  Hence the name: one frequency is an alias of the other.
> 
> And since they're indistinguishable, aliasing components occur at the same volume level as the intended audio components, so that's really bad.
> 
> That mechanism doesn't happen in magnetic tape.  In tape there is some difference frequency content, but it's at a much lower level since it's a side effect of the nonlinearities.
> 
> Refs:
>    High Frequency Bias Requirement for Magnetic Tape Recording, 3M
>    http://www.aes.org/aeshc/docs/3mtape/soundtalk/soundtalkv1n2.pdf
> 
>    Biasing in Magnetic Tape Recording, Ampex
>    http://thehistoryofrecording.com/Papers/Jay_McKnight/Biasing_in_Magnetic_Tape_Recording.pdf
> 
>  -- Don
> 
> --
> Don Tillman
> Palo Alto, California
> don at till.com
> http://www.till.com
> 
> 
>> On Jan 28, 2016, at 7:46 AM, nvawter <nvawter at media.mit.edu> wrote:
>> 
>> holy cow, that's awesome :)
>> 
>> It's begging to be exploited!  I wonder what it would be like to make lo-fi bias signals?  Low-frequency?  non-sinusoidal?  wow and flutter in the bias signal?
>> 
>>> On 2016-01-28 10:38, Sarah Thompson wrote:
>>> It seems to me that nobody ever really acknowledges that analog tape
>>> actually also samples the recorded signal in time, though doesn't
>>> quantize in level. The bias signal, usually a sine wave at 50 - 100KHz
>>> or so, is added to preemphasized version of the audio as it is fed to
>>> the record head. This overcomes the large amount of hysteresis in the
>>> magnetic materials in the tape by rapidly flipping between linear(ish)
>>> regions. This overcomes the coercivity of the iron oxide, but has the
>>> side effect that the audio is only really recorded during the peaks of
>>> the bias signal. Though not identical to digital sampling, this does
>>> still cause very similar aliasing issues, so the signal path still
>>> needs antialiasing filters, even though they are typically not thought
>>> of as such.
>>> This is also why higher bias frequencies often sound better, though
>>> were harder to achieve due to the relatively high voltages needed.
> _______________________________________________
> Synth-diy mailing list
> Synth-diy at dropmix.xs4all.nl
> http://dropmix.xs4all.nl/mailman/listinfo/synth-diy