In article <CANJYbi8=W6Z5doDVQB4gy2rLC9kA9pkTPE49AszWQMDgTsC+HQ at mail.gmail.com>, Wolfgang Eichberger <oe5ewl at gmail.com> writes: I haven't investigated to know for sure, but I think the Aesthedes design station has these for the disks. (I doubt anyone on this list has one of these; there is one other in the Netherlands that I know of.) Tektronix 4170 local graphics processor. -- "The Direct3D Graphics Pipeline" -- DirectX 9 version available for download <http://legalizeadulthood.wordpress.com/the-direct3d-graphics-pipeline/> Legalize Adulthood! <http://legalizeadulthood.wordpress.com>

In article <alpine.DEB.2.02.1111301207500.9421 at linuxserv.home>, Christian Corti <cc at informatik.uni-stuttgart.de> writes: This is confirmed by <http://bitsavers.org/pdf/seagate/ST506_Preliminary_OEM_Manual_Apr81.pdf> See pages 20-22. -- "The Direct3D Graphics Pipeline" -- DirectX 9 version available for download <http://legalizeadulthood.wordpress.com/the-direct3d-graphics-pipeline/> Legalize Adulthood! <http://legalizeadulthood.wordpress.com>

On 2011 Nov 30, at 12:27 PM, Tony Duell wrote: Not really. The pulses or flux transitions from the disk are in discrete time slots relative to each other, albeit with some fluctuation due to physical reality, but that fluctuation is not part of the intended information content. It is not like pulse-position-modulation or pulse-width-modulation where the intended information is represented by a continuous value ('analog') of the time between two events (edges). The disc-ferret/cat-weasel/etc. could be said to do a digitally- sampled-analog-interpretation of those discrete time slots for the sake of helping with data interpretation and recovery.

On 2011 Dec 1, at 12:12 PM, David Riley wrote: (As others have mentioned (including Tony), it is quantized in amplitude (to two states) by drive read circuitry.) In time, it was quantized when it was recorded and retains that quantisation when read (note I said discrete time slots relative to each other). You might also note the part of my message you cut in your reply: "The disc-ferret/cat-weasel/etc. could be said to do a digitally- sampled-analog-interpretation of those discrete time slots for the sake of helping with data interpretation and recovery." One can do an analog assessment of a digital logic voltage level (e.g. 3.00V vs 3.02V) but that doesn't make the signal analog. While there is some room for a little looseness in phrasing when dealing with (signals from) a physical medium, it's still a stretch to call the disk signal 'analog in the time domain', esp. when the phrase has a firmer meaning in other applications.

On 2011 Dec 1, at 1:23 PM, Eric Smith wrote: (In reply to this and your previous message) I'd differ on that: it is still time-quantized on read even if the timing isn't quite the same as when it was written. The raw read data still comes in 'time chunks', able to be directly correlated to the raw written data. The data may not be quite the way you want it (it includes clock transitions) and you need a transformation to get it to the way you want, but that transformation is (or primarily is) a discrete process or transformation. That it is possible to produce a data-separator that recovers the data so reliably is exactly the point. (Perhaps we have different concepts of what 'time-quantised' or simply 'quantised' means. While there are differences, one might ask the same question of an async serial signal for a simpler case to examine.)

On Dec 1, 2011, at 5:58 PM, Eric Smith wrote: I would second this description wholeheartedly. What the DiscFerret and similar devices are doing is *re-quantizing* the data which has been *de-quantized* (i.e. randomly perturbed) by the drive mechanism and other random factors (including things like the differences between the controller that wrote the drive with arbitrary base frequency and drift). After the signal has been re-quantized, it often needs significant work on it to re-align it in the time domain with the expected clock (thus the separator). - Dave

Brent Hilpert wrote: continuously through signals being In terms of modulation complexity, yes. (demodulation), finite number relations At the transmit end. Not by the time it goes through the phone system (including parts of the modems such as the hybrid). Just as I've been saying about magnetic disk, the channel changes the nature of the signal dramatically, such that it is actually extremely difficult to quantize it correctly at the receiving end. multi-bit bauds). Symbols (information-theoretic) or constellation points (modulation terminology). don't consider to the receiving Yes. For amplitude quantization, if the transmitter and channel can guarantee (for example) that a signal is always under 0.8V for a zero and over 2.0V for a one, at the receiving end, then it is already quantized. If any non-trivial processing is necessary to recover what the transmit amplitude was, then the signal was not amplitude-quantized prior to that processing. In the time domain, if the transmitter and channel can provide the receiver with a simple unambiguous indication of the correct times to sample the data, it is time quantized. If any non-trivial processing is necessary to recover that timing information, then the signal is not time-quantized prior to that processing. represented in the time chunks, consideration of the I would only say that that indicates that there is encoded digital data present, but not that it is quantized at the point where it is received. It is the job of the receiver to quantize it back into the original digital data. noise/perturbations/fluctuations/deformation de-quantize the signal If there were hard, quantifiable limits, I might agree with you. However, that's not how it works. For timing of magnetic disk flux transitions, and for all aspects of V.32 modulation, as you get further point where the signal was a 0 to a point where the signal was a 1. There is a considerable overlap region of uncertainty. When the system is working well, the signal doesn't spend much time in that region. signal Of course. With TTL signals, there are known simple design criteria which can be used to ensure that the signals make it from the driver to the receiver without violating the limits on logic level margins and without significant timing distortion. With magnetic media and with V.32 modulation, the same cannot be said. (sometimes doesn't disqualify I guess we'll have to agree to disagree on that. In my opinion, if it is quantized, there is a trivial and entirely unambiguous interpretation of the signal, and with complex modulations and noisy channels, that clearly isn't the case. I know that the modem engineers I worked with did not consider the analog signal arriving at the ADC in the modem to be quantized. They only considered the data to be quantized after they sampled the ADC, hilbert-transformed it with digital filters, and had a carrier tracking loop determine the times to sample the result of the filters. Before that point, it was considered a noisy, non-quantized signal.

Tony Duell wrote (about the read data signals of the ST506/412 drive interface): Brent Hilpert wrote: discrete time slots relative to each other, When writing, sure. fluctuation is not part of the intended information content. No, but it's part of what you actually get when you read data from the drive, so therefore the raw read data at the interface is not time-quantized, which is what Tony said. It is part of the job of the data separator to time-quantize it back to discrete bit (and clock windows), and doing a good job of that is extremely non-trivial. What *is* trivial is to time-quantize it to some window that is significantly smaller than the bit window (i.e., higher sampling frequency), which is what the Catweasel and Diskferret do. At that point it is time-quantized but not fit into bit windows, so data separation is still required.

On 2011 Dec 2, at 2:38 PM, Tony Duell wrote: I think we all agree that the 'analog' assessment (more accurately fine-grained discrete sampling) performed by a disk reader/archiver of the transition timing of the signal from the disk is useful. That one can record transitions onto the disk more or less arbitrarily in time, that there may be multiple recording formats, is another motivation - as you say - for the analog assessment. In my response however, I was distinguishing between the assessment of the signal and the properties of the signal, as your response had alluded to the latter, although we also have the two perspectives to consider of the agglomeration of possible signals from all disks versus a given instance. Leaving aside the low probability or occurrence of someone using the disk for analog PWM or PPM recording, and other opinions notwithstanding (other thread), I do not consider a given instance of a (typical) amplitude-normalised (to two states) read signal from the disk to be analog in the time domain, but from your other message I think we agree on that. In the analogy with the FSK modem in your other message then: simply put, whether the demodulated signal is time-quantised depends on the properties of the modulating signal.

On 2011 Dec 3, at 12:57 AM, Eric Smith wrote: I'm not sure what you mean by "representation of the signal", although I was going to posit a similar distinction to you. I presume you mean the phenomena - changing amplitude over time of an electrical 'signal' (term used loosely) for instance. However a 'signal' (term used more strictly) is something intended to convey information (content), and generally there is some prior knowledge involved of how to interpret the phenomena to retrieve that information. I would say that conveying discrete information through a medium or channel is typically done with a quantised signal. The original discussion was for a baseband/pulse signal though, the addition of modulated-carrier mediums was an additional complexity that I didn't feel was warranted for the discussion and brings in additional considerations. This does seem like one of those discussions that would be sorted out in a minute of conversation in front of a blackboard, but over email - with slightly different understandings or ambiguous use of words - it ends up in an infinite regression. I'm not sure of the continuing relevance to the original discussion but we'll see where your thought experiment goes. I expect we agree it is not. It is a discretely-rendered approximation of the sound. I expect we agree it quantised. Not likely by the time it's emanating from the speaker. If one were looking at the output of the DAC, however, I would say it was quantised if the discrete steps are still discernable in both time and amplitude and one could precisely recreate the input to the DAC, for any arbitrary input sequence (assumes a good DAC). Once it has gone through the [limited-bandwidth,distorting,noisy] filters, amplifier and speaker, info has been lost and you're not going to reliably or consistently get back to arbitrary 96KHz, 24-bit discrete data. And we both know (my) answer to that, I expect we agree on this question.

On 2011 Dec 3, at 2:03 PM, Tony Duell wrote: Well, I wasn't really thinking of the controller, rather the signal one typically sees, which - if you like - implies a controller, as something wrote the original signal. The original context of the thread was recovering digital data from an ST-506 drive, albeit one with unknown controller characteristics, "Waveforms" were mentioned. I allowed for the benefit of using an analog-in-the-time-domain assessment of the signal from the drive in my first message. (And here I thought we would have agreed, based on your assessment of the async serial signal vs. cat notifier). Why would I argue as you suggest? It is a question which does not follow from my statement. Your question which I was answering referred to the "receive data output" of the modem. If there is a debate still it seems to be one of semantics. Is a wire interconnecting a gate output and gate input of two TTL ICs in a synchronous system analog? Or is it an analog *link*? Or is it just *capable* - if extracted for another use - of being an analog link? Or does it just have some analog characteristics? Or does it merely have the capability of *conveying* analog information when used in an other context? One could answer yes to all those, depending on the nuance one wishes to put on it. While it can be useful and necessary to analyse the signal on that wire in analog terms at times (prototyping, repairing), do we normally think of it or treat it as analog? I didn't think so: because we know something about the characteristics of the signal we expect to see on it.

On 29 Nov 2011 at 23:18, Philip Pemberton wrote: This is where I wish we had teleportation--I've got stacks of various old ST506-interface controllers and drives. One I don't have is an ARLL Perstor controller--that might be interesting. Phil, what are you using for your sample rate? I'm thinking that 30MHz might be adequate for the job, since hard drives don't seem to have the ISV issues that floppies do. --Chuck

In article <4ED675E6.3080206 at philpem.me.uk>, Philip Pemberton <classiccmp at philpem.me.uk> writes: Since ST-506 is a digital interface, is there any point to digitizing the digital signals as if they were analog signals? -- "The Direct3D Graphics Pipeline" -- DirectX 9 version available for download <http://legalizeadulthood.wordpress.com/the-direct3d-graphics-pipeline/> Legalize Adulthood! <http://legalizeadulthood.wordpress.com>

On 30 Nov 2011 at 11:38, Richard wrote: Sorry for the confusion. I was referring to the acquisition rate. In other words, did a read pulse occur in this 33nsec window (30MHz)? Recording each even in terms of ticks of a 30MHz clock from the previous event.. Chuck

On 30 Nov 2011 at 22:30, Philip Pemberton wrote: Exactly what I was (trying to say)/saying. At 30 MHz every 33nsec window is used to sample for a pulse occurring within the window. You run a counter that counts between such transitions--and store the accumulated count when a transition happens. What the catweasel and countless others have been doing for years. I suspect that the Kyroflux and Deviceside devices do likewise. It can be done with almost any modern microcontroller; no fancy FPGAs needed as most modern microcontrollers contain counters with a "capture" facility triggered automatically. For example, a lowly ATMega running at 16MHz is more than sufficient to get accurate samples to decode a high-density floppy. Similarly, the PWM facility on most microcontrollers is sufficient to simulate a high-density floppy. All you need is a counter, some way to trigger it and some memory to store the counts in. It's not rocket science--in fact, it's mostly software. My apologies if I'm not communicating well. --Chuck