| |
Hoogo
Registered: Jun 2002
Posts: 105 |
scope view of some tape signals
I've connected the scope to the r/w head of the Datasette and made some tests. The test signals were coming from the C64 and recorded on a Datasette with a normal Type I tape.
Most test signals were made of "bits" of a length of 22 cycles. That length come close to the usual sample rate of 44100Hz, might be handy some day to create files with a sound card.
Also the test signals were really simple until now, always the same peak and gap lengths repeated.
------------------------------
Upper row:
-Noise: I just pressed the scope to the head without pressing play. I assume that the signal comes from the monitor, so one wave should be 63 cycles. I don't see that noise in the other signals, so I will better check again what I did there.
-44+44: 11.025 Hz, somewhere near the maximum frequency of a tape. The signal is not stronger than the noise, imho not usable for any turbo tape.
-55+55 can't be created that symmetric with a sound card, . Seems clear enough for a tape.
-The last 3 in the row were tests with an assymmetric signal, smaller peaks and longer gaps. Surprisingly it looks just like the normal sinus of weaker level.
------------------------------
Lower row:
-Nothing (=the noise of the scope?)
-The normal tape format, just saved something with the save command.
-Something very long. I just added a loop from $40.
-Something VERY long, loop from $00. The scope was adjusted to the longer signals. Apparently the signal falls to the base line when it gets too long.
-And finally: My finger pressed to the head. As this is perfectly one pulse per line, I'm confident that this signal comes from the monitor.
The signal should be steep when it crosses the baseline, so the long flattening signals should be worse than the fast signals.
------------------------------
Coming next, maybe this weekend:
-Check the noise again.
-See the writing signal at the head.
-Compare the input received from the r/w head to the resulting signal on the tape port. Unfortunately there's only one sensor attached to the scope.
-See more complicated signals made from different peaks and gaps
-Write quick changes at the center of a peak
-See a chrome tape
-See something written by Hifi equipment. |
|
| |
Xiny6581
Registered: Feb 2004
Posts: 73 |
As a preserver those kind of "tests" always bringing a smile upon my face :)
Way to go and make more of them, with rich documentations.
Keep it up dude ;) |
| |
ChristopherJam
Registered: Aug 2004
Posts: 1409 |
Those are really interesting, thanks Hoogo!
The latter three on the top row all looking the same is consistent with any components at 11HKz or higher being lost altogether; the duty cycle is contained in the higher harmonics. Given the measurements you've shown, I'd expect something like the following for 44/132, 55/165, 66/198:
Also, would you mind measuring a digital signal output from the c64? (eg from the serial port or user port)
I'd be curious to see how (eg) 44 cycles on/44 cycles off looks with your measuring gear, to see if the scope itself is bandwidth limiting the input at all. |
| |
Hoogo
Registered: Jun 2002
Posts: 105 |
Quoting ChristopherJam...I'd expect something like the following for 44/132, 55/165, 66/198:
...
Also, would you mind measuring a digital signal output from the c64? (eg from the serial port or user port). I also hope to see some flattened peaks somehow. But I wonder how the phase of the higher frequency is generated? And shouldn't the 110+66 signal already show such an effect?
I'll test the cassette write, with and without attached Datasette. |
| |
ChristopherJam
Registered: Aug 2004
Posts: 1409 |
Quoting HoogoI also hope to see some flattened peaks somehow. But I wonder how the phase of the higher frequency is generated? And shouldn't the 110+66 signal already show such an effect?
I've not derived a close form equation for the higher frequency amplitude or phases; the graphs above just came from applying a low pass filter that roughly matched the behaviour of your first few graphs (zero response at T=88 cycles or less, pass through at T=176 cycles or more).
The 110+66 signal has a period of 176 cycles (frequency 5.6KHz), so the second harmonic has a period of just 88 (frequency 11.2KHz). Given the negligible signal you get back for 44+44 I wouldn't expect to see anything other than the fundamental for 110+66.
Quote:I'll test the cassette write, with and without attached Datasette.
Ah! Excellent plan. |
| |
Hoogo
Registered: Jun 2002
Posts: 105 |
Unfortunately I changed the timebase of the scope quite often, so the faster signals may look wider than the slow ones.
-----------
Upper rows:
The signal on the r/w head while writing.
Those are 100 times stronger than the read signals. Maybe a bandpass of the 1st derivation of such a written signal would come close to the reading?
The last 2 in the upper row were 2 experiments to shape the output.
The first was 66 cycles for each peak/gap. The first 33 cycles were spent for 11 quick changes between 0/1, the other 33 were written normally.
The 2nd was 63 cycles, 21 normal, a fast mix in the center, then the normal again.
The result in the medium row show no different shape at all.
----------
Lower row:
-The signal on the write line on the port. It made no difference if the Datasette was attached or not.
-The noise again, looks less sinus on this picture now. Maybe the cable of the sensor of the scope has a problem.
-The readings of the 2 asymetric signals you suggested. |
| |
Hoogo
Registered: Jun 2002
Posts: 105 |
Some more test pictures. Base length for all was 22 cycles again.
----------------------------------------
Upper 2 rows are combinations of a few different lengths, 2 timeframes, written signal an read signal. First number is Bit written as 1, next one is written as 0. Alltogether 16 bits * 22 cycles=352 cycles.
-The written lengths can be found (more or less and at least here...) as distant between the peaks in the read signal.
-The distant measured by the C64 should be one crossing of the zero line, and that distance is different from the written one.
----------------------------------------
The lower row shows some more weird results
-16 Bits should be filled with 4 waves of size 2+2, but I count 5.5
-The written distances don't seem to match the peaks in the result.
-Not every wave crosses the zero line, one written bit pair will be invisible to C64.
-A single written bit of 22 cycles is really, really short, but the created peaks look clearer to me than the small peaks in the previous picture, even though those were made with longer bits.
-A chromium tape. Very noisy, mabe some music that wasn't deleted?
----------------------------------------
Next I tested the jitter.
I've chosen 88+88=176 cycles. That signal is in one of the older pictures and comes close to typical short bits of some Turbos.
measured cycles / count
========
73 1
167 1
171 4
173 196
175 768
177 511
179 47
181 1
199 1
========
The outliers were measured in different runs.
The jitter is imho great, I assume the peak would even be clearer if I had wasted another cycle to hit the even cycles.
----------------------------------------
Conclusions and thoughts so far:
-It's not enough to care for an exact timing while writing the tape. Same lengths written can result in very different peaks read.
-I think it's neccessary to know the previously written bits to find the perfect moment to flip a bit and create a nice timing for reading.
-I think now that the heavy jitter that Turbos have to handle is not caused by bad tape mechanics, but by a "bad" timing while writing.
Understanding more of it would be helpful.
Wikipedia has a nice picture of voltage and current of a coil:
I guess the current comes close to the magnetic field that is stored on tape, and reading is a bigger window on that tape, and the final result is a derivation of the average field in that window... But that's to be discussed. |
| |
chatGPZ
Registered: Dec 2001
Posts: 11386 |
Quote:I think it's neccessary to know the previously written bits to find the perfect moment to flip a bit and create a nice timing for reading.
what you are looking for is "precompensation". you might want to look at the datasheets of some WD floppy controller chips, i remember it being described in detail somewhere. The same effect likely exists on tapes (and any other magnetic media) |
| |
tlr
Registered: Sep 2003
Posts: 1790 |
For how the data is actually recorded onto the tape is not enough just to consider the recording head coil. There is something called magnetic hysteresis (in the tape itself) and that needs to be compensated for, at least if you are trying to record something linearly like sound. Typically recording equipment employs a bias tone super imposed on to the signal.
The requirement on this differs between tape types so the recording deck needs to have specific support for the type of tape used, normal, CrO2 or Metal. The playback deck does not however. |
| |
chatGPZ
Registered: Dec 2001
Posts: 11386 |
not sure of that plays a big role for the c2n, since here the tape is recorded fully saturated (which never happens with audio) |
| |
tlr
Registered: Sep 2003
Posts: 1790 |
Maybe not for the actual read binary data but for understanding how a particular current curve through the recording head ends up as magnetisation on the tape (as the post speculated about). |
| |
chatGPZ
Registered: Dec 2001
Posts: 11386 |
oh, yes, of course. it also helps to understand what precompensation does, and why. |
| |
tlr
Registered: Sep 2003
Posts: 1790 |
IIRC the effect that needs precompensation only occurs when the information density gets sufficiently close to the particle size of the medium like in a HDD or floppy. I wouldn't think it applies to the rates applied to a typical tape, especially since the recorded bit rate is far below the audio bandwidth of a typical cassette tape.
What I think could apply though is the "memory" effect of the simple drive circuit, being a coil driven by a square wave through a resistor. Maybe that is what hoogo was refering to with the graph?
I don't know the inductance of the recording head so I can't calculate the time constant here, but the drive resistor is 10k in the c2n schematic. |
