Maximum bitrate for storing information on home-quality cassette tapes?

[MauriceH]

Oww Fond some hole in Google to dive into.
TEAC R-80 , 4 chan Data Recorder (E-BAy)
Teac R-71 (E-Bay)
Then found the Manual TEAC R-60, looks like it uses FM-Mod.

So in the end, there seems to be a lot of trial and error to push limits of a standard Audio-Cassette.
Think all little steps to the next level real Data-Cassette Tape Storage units.
As of course from early 60's the reel-to-reel we all know very well.
This has to do with the low-cost data storage.

 

[cj7hawk]

Those "computer program data" cassette recorders were just standard dictation-style mono cassette recorders, with few if any modifications. Some simply have the buttons and indicators relabeled for computer use (i.e. "load" instead of "play", "save" instead of "record"). Some have a monitor you can switch on to hear what is being output from the computer, helpful for computers like the TRS-80 which can also play music and sound effects through the cassette port. Some have a preset volume level that is supposed to be matched to the computer's requirement. That Philips is the first I've seen to also have a normal/reverse phase switch, which in some fringe cases can make a difference, but usually doesn't.

The National (Panasonic) RQ-8100 also has the phase switch. I managed to find one of those, but it doesn't record - I think the head is damaged. Not sure how that happens to a cassette recorder though. The signal makes it to the head, but nothing is recorded.

Anyway, I never figured out why they put that there, or what it does. I can only assume that it's related to some specific early PC that needed the initial mark or start bit to have a specific orientation like it was using some kind of group code? But I have no idea of any such recording scheme existed in this context.

 

[cj7hawk]

So in the end, there seems to be a lot of trial and error to push limits of a standard Audio-Cassette.
Think all little steps to the next level real Data-Cassette Tape Storage units.
As of course from early 60's the reel-to-reel we all know very well.
This has to do with the low-cost data storage.

There was definitely a lot of research here, and would have been interest back in the 80s, but computers moved very quickly into disk drives... after that they were solving a problem that had already been solved.

I don't understand a lot about the tapes, but what is obvious is that tape recorders had a much higher bandwidth than phone lines (POTS) and phone lines managed to handle 56K modems pretty early on.

They had some pretty complex encoding methods though, but if a phone line can handle it, then a tape recorder should be able to.

Even a 64 Kbps line can handle 9600 baud for faxes, and a 16kbps line can handle 1200 baud.

 

[MauriceH]

"...I think the head is damaged. Not sure how that happens to a cassette recorder though..."
Look at your erase head, most likely that has problems with the glue that holds that ferrite/ coil in its housing,
Beware you can't just put an other in, has to be right inductance.

You see right side already lost a bit of ferrite and its sticky due to degrading glue.
If it looks ok, no sticky, probably just clean the rec/play switch.

 

[cj7hawk]

"...I think the head is damaged. Not sure how that happens to a cassette recorder though..."
Look at your erase head, most likely that has problems with the glue that holds that ferrite/ coil in its housing,
Beware you can't just put an other in, has to be right inductance.

You see right side already lost a bit of ferrite and its sticky due to degrading glue.
If it looks ok, no sticky, probably just clean the rec/play switch.

The erase head is passive, just a magnet that it moves to be in contact with the tape when recording, and the read/write head has no resistance between the two pins - it's open circuit. One side does have resistance to the case ( and GND) though.

 

[cjs]

Anyway, I never figured out why they put that there, or what it does. I can only assume that it's related to some specific early PC that needed the initial mark or start bit to have a specific orientation like it was using some kind of group code? But I have no idea of any such recording scheme existed in this context.

What it does is swap the phase: when it sees a flux transition that would give you a positive signal, the signal goes negative instead, and vice versa.

It's fairly common on Japanese CMT decks designed for computer use. I don't now recall exactly which systems needed it, or where it made a difference, though. (My impression was that most systems counted nominal zero crossings of an AC signal, and so this would make little difference though, come to think of it, it could made a difference on certain high-speed signals where you have a single cycle of a waveform indicating a mark, space or similar.)

I don't understand a lot about the tapes, but what is obvious is that tape recorders had a much higher bandwidth than phone lines (POTS) and phone lines managed to handle 56K modems pretty early on.

56K modems didn't appear until a couple of decades after people started using CMT data storage, and they only ever worked on digital systems. (These were 56K in only one direction, and the modems themselves were entirely digital, with an ISDN PRI or similar connection to the network.)

Even leaving aside the processing power required to do analogue 19.2k or 28.8k, the phone network differs in some important respects from CMT, particularly in that when you send a frequency on one side, that exact frequency comes out the other, not something played back slightly faster or slower, much less changing speed over time.

The erase head is passive, just a magnet that it moves to be in contact with the tape when recording...

This would be unusual; typically erase heads are similar to the record/play heads except that they are energised with a constant frequency signal that is high enough that it leaves the tape in an "erased" state, usually the bias signal I believe. (That said, apparently permanent magnet erase heads do exist.)

 

[cj7hawk]

What it does is swap the phase: when it sees a flux transition that would give you a positive signal, the signal goes negative instead, and vice versa.

It's fairly common on Japanese CMT decks designed for computer use. I don't now recall exactly which systems needed it, or where it made a difference, though. (My impression was that most systems counted nominal zero crossings of an AC signal, and so this would make little difference though, come to think of it, it could made a difference on certain high-speed signals where you have a single cycle of a waveform indicating a mark, space or similar.)

If you ever do come across a system that needs this, I'd be curious about it- so please let me know.

56K modems didn't appear until a couple of decades after people started using CMT data storage, and they only ever worked on digital systems. (These were 56K in only one direction, and the modems themselves were entirely digital, with an ISDN PRI or similar connection to the network.)

At the ISP end they were digital, but analog at the CPE (Customer) side. You could run them analog both ways, but then you got 33Kbps full-duplex rather than 56K.

Still, it did talk well of the capacity of the phone lines. Even back then.

BTW, I didn't know they had PRI based modems you could run at home, which I think is what you're suggesting? Generally if you had a PRI, you just connected digitally on two B channels and got 128K here - or 108K in the US IIRC from my worldcom days. And there was a 16Kbps D channel as well. But the modems at the ISP itself generally got a T1 or E1 feeding into them, running some interesting L2 protocol like NRZ or HDB3. I never did much with them - by the time I was working with telco stuff, Western Australia was rapidly shifting to fiber and Ethernet as a carriage protocol at layer 2 rather than SDH or PDH, and ATM never really took off here at all. That happened in my part of the world around 1996, which was maybe a decade ahead of the rest of Australia - mainly due to our isolation, so the local telcos did whatever they wanted, and people wanted data, so they just sold ethernet everywhere and ran networks off of commercial swiches. It sounds rather poor, but it was exceptional - suddently gigabit got a lot cheaper and you could get dark fiber at the cost of laying it in the ground.

My city probably still has a lot of unused and lost dark fiber from that era, most of it still probably owned by the government. I imagine some day someone will just start using it.

Even leaving aside the processing power required to do analogue 19.2k or 28.8k, the phone network differs in some important respects from CMT, particularly in that when you send a frequency on one side, that exact frequency comes out the other, not something played back slightly faster or slower, much less changing speed over time.

Agreed, hence the need to recover clock with the signal, for which FM at least is well suited, and theoretically you could change the speed by any amount during transmission without loss allowing for dynamically being able to change the rate of data during communications - eg, sliding from 1Kbps to 16Kbps then back to 1kbps would sound like a kids whistle going up and down... I previously directly decoded information from a magstripe reader like that, the kind you push through by hand - and I just synced to the sync pulses then adjusted the timerate based on the last edge received. It worked pretty well and didn't mind that people aren't all that consistent in pulling the card through. Modern readers do it automatically... But early readers were rather primitive and only communicated the edges.

This would be unusual; typically erase heads are similar to the record/play heads except that they are energised with a constant frequency signal that is high enough that it leaves the tape in an "erased" state, usually the bias signal I believe. (That said, apparently permanent magnet erase heads do exist.)

Yes, it was a surprise to me also when I noticed that there were no wires going to the erase head. Of course, it wasn't recording, so I couldn't validate that it was erasing... It looks like a normal erase head from the front, except that there's nothing connected to it. I didn't even know they existed - I mean, it makes sense, but as a kid I tried to erase a cassette with a magnet but it didn't work. I guess the "gap" is needed to create the correct field. I'd love to get a working one of this model, but they only tend to show up in Japan. We got a lot of cheap Sharp stuff here in Australia, but I never saw one of these National ones before with the phase selector.

 

[jlang]

The National (Panasonic) RQ-8100 also has the phase switch.

Anyway, I never figured out why they put that there, or what it does. I can only assume that it's related to some specific early PC that needed the initial mark or start bit to have a specific orientation like it was using some kind of group code? But I have no idea of any such recording scheme existed in this context.

The phase switch would be useful for phase encoded data. The Tarbell s-100 cassette board would be one example.
The board had jumpers to flip the phase of the data out and data in.

 

[cj7hawk]

Does anyone know what this is? Seems awefully a lot like what we've been talking about... And looks very old.

On ebay presently.

 

[cj7hawk]

 

[krebizfan]

It uses RS-232 serial ports. The advertised (intended) use would be as a tape recorder alongside a serial terminal replacing paper tape as local storage. Record the data or play it back.

Wonderfully overengineered.

 

[cj7hawk]

It uses RS-232 serial ports. The advertised (intended) use would be as a tape recorder alongside a serial terminal replacing paper tape as local storage. Record the data or play it back.

Wonderfully overengineered.

Indeed. It looks like it goes up to 9600 Baud, though while it looks like a normal audio tape, there's no certainty around the heads or other mechanisms inside.

 

[krebizfan]

Note the notch on the tape. It is a streaming tape. Not sure which streaming tape variation but it was probably fairly early and low performing. Techtran ran ads from about 1976 selling basically the same device in several different cases for close to a decade.

 

[MauriceH]

If you blow up the picture of the cassette the answer is there.
The Rolm CBX Telephone system

So think its a service kit to program those systems.

The cassette:
ROLM CBX SYSTEM
HTYTEL5280.CS
1200 BAUD 1 of 3
17feb1983 -REL 7

 

[mark0x01]

The cassette looks similar to the Teac data tapes, shown here next to a standard audio tape.

 

[cj7hawk]

The cassette looks similar to the Teac data tapes, shown here next to a standard audio tape.

Well, in fairness, they both look like normal audio cassettes more than they look like each other... But it is possible - likely even - that whatever recording system they use is not even remotely connected to how audio cassettes work.

 

[lowen]

I was just wondering what the top practical speed is for storing information on normal audio cassettes.

Ok, so a typical audio compact cassette will have a head gap of 2 μm and a tape speed of 1-7/8 IPS, or 47,625 μm per second, or 23,812.5 gaps per second, giving a max recordable frequency of half that, or 11,906.25Hz. Let's round that to 12kHz and call it a day.

Tape can record analog values in each magnetic domain, so you're not limited to just saturated-unsaturated digital recording, but different recorders and different types have different dynamic ranges; I've seen a high end spec of over 65dB, but a typical cassette is going to be hard pressed to go past 50dB.

So, your symbol rate can be up to 12 kilobaud, but you could easily have 8-bit symbols encoded as voltage levels and phasing ((2^bits-1) = dynamic range; dynamic range for 8 bits is 255; that's 48.1 dB (voltage, not power, ratio, so 20log(x) and not 10log(x) ), which fits in the typical 50dB of a cassette.

In practice, such a 256-QAM system running at 96kbps would likely have pretty high BER due to other signal imperfections in the cassette system, such as wow and flutter, and gain variations. But some other multiple of 12kHz could be managed.

Analog tape is designed for as close to a linear response as possible, with appropriate bias to deal with hysteresis. So analog techniques are required.

 

[cj7hawk]

Ok, so a typical audio compact cassette will have a head gap of 2 μm and a tape speed of 1-7/8 IPS, or 47,625 μm per second, or 23,812.5 gaps per second, giving a max recordable frequency of half that, or 11,906.25Hz. Let's round that to 12kHz and call it a day.

Tape can record analog values in each magnetic domain, so you're not limited to just saturated-unsaturated digital recording, but different recorders and different types have different dynamic ranges; I've seen a high end spec of over 65dB, but a typical cassette is going to be hard pressed to go past 50dB.

So, your symbol rate can be up to 12 kilobaud, but you could easily have 8-bit symbols encoded as voltage levels and phasing. In practice, such a 256-QAM system running at 96kbps would likely have pretty high BER due to other signal imperfections in the cassette system, such as wow and flutter, and gain variations. But some other multiple of 12kHz could be managed.

Analog tape is designed for as close to a linear response as possible, with appropriate bias to deal with hysteresis. So analog techniques are required.

Wow ! Did you just look that up or was that all off the top of your head ! That's some impressive detail you got in there - and it makes sense from a hearing perspective from what I can see.

I did some experiments with digital input on the record line and noted that the recovered signal from the ear socket was particularly sensitive to frequency with respect to the recovered amplitude, indicating that a good recovery system would need it's own amplifier and an op-amp or comparator at the minumum to recover a signal, and had no problem getting edges up to 10KHz, but the amplitude was hugely affected by the frequency with digital signals recoverable cleanly up to about 2 KHz, and from there it dropped rapidly - I never tested past 10 KHz, as I was wondering if the noise floor was going to interfere with the signal, but at lest on the same player I got consistent edges to near microsecond level - and my scope didn't jump around at all suggesting it was pretty consistent - way more than I see on some disk drives.

Though I haven't tested to see if encoding effects that a lot - which I suspect it will, with the constantly changing frequency, since that affects the amplitude, which in turn will cause micro-DC offset switching between ones and zeroes.

I need to make up some circuitry and see what happens when I do.

I thought about QAM-256 type schemes, but they don't generally cope well with speed variations and I don't have a lot of data on the range of variation that can be expected, though I expect it should be small. Interestingly, I remember large flywheels on older tape systems that the small "walkman-style" cheap tape system I am testing with doesn't seem like it would have much of one there.

Still though - 96 Kbps data rate would be crazy fast ! That would eat floppy disks alive for loading of linear code - such as games. It could fill 256K of memory in just about 30 seconds.

I was thinking that encoding something at a slightly higher rate than a standard baud rate would make it possible to do things like use a UART to send/receive the information, and would let you do things like include parity in recordings at the byte level - while adjusting the speed of the rate at the bit level rather than at the byte level as most asynchronous UARTS tend to do, so you would get an improvement in the WOW resilience due to the bit-level timing variation resilience, but you'd need a start bit for every byte at the minimum, and functionally you're running at two rates - The tape bit rate and the UART bit rate which complicates things slightly, but asynchronous protocols would handle it well.

If 12KHz is a symbol rate, with each symbol having two transitions then MFM can at least get 24 kbps with a max loss of timing spaced over two bits. Also some thought is needed as to whether and how to encode blank periods between bytes.

Thank you for the added information - there's a lot there I hadn't thought about or didn't know.

 

[lowen]

Wow ! Did you just look that up or was that all off the top of your head ! That's some impressive detail you got in there - and it makes sense from a hearing perspective from what I can see.

A combination; I needed to remind myself of the gap size, but inch and seven-eighths is off the top of my head; I've had an on again off again side gig as a broadcast engineer for thirty five years. 15, 7.5, 3.75, 1.875 - the natural progression of tape speeds. 15 IPS reel; 7.5 IPS Fidelipac/NAB cart; 3.75 IPS eight-track; 1.875 cassette.

I did have a couple of cassette systems that could do 3.75 IPS, though: a Marantz dual speed and a Tascam portastudio 246 four track. With a good Type II tape that portastudio 246 could do 20Hz-18kHz at 3.75 IPS, 40Hz-14kHz +/- 3dB; at 1.875 the specs are really good, 40Hz-14kHz and 40Hz-12.5kHz+/- 3dB. Bias frequency 85kHz in that deck. Must have had a narrower than normal gap.

The rest is pretty simple math and Nyquist.

I did some experiments ...

It's an interesting project, for sure. Using really high quality C-60 tapes with some of the later superferric formulations with insanely high MOL, SOL, and dynamic range numbers will make a difference. Using professional grade transports (Tascam 112, Marantz PMD 502, or similar) that have been properly refurbished (I reworked my share of both of those; the 112 was pretty easy to work on) will make a huge difference to the maximum data rate.

If 12KHz is a symbol rate, with each symbol having two transitions then MFM can at least get 24 kbps with a max loss of timing spaced over two bits. Also some thought is needed as to whether and how to encode blank periods between bytes.

12kHz is the Nyquist sampling theorem frequency; the magnetic flux transition 'sampling' rate is limited by the gap size to 24k transitions per second. This isn't a hard limit; there are aliasing artifacts as you approach and exceed the Nyquist frequency. You could have a symbol be a voltage level but not an extra transition, as that takes you over the Nyquist frequency and you get aliasing. The phase accuracy suffers the closer to the Nyquist frequency as well, and so symbols where the amplitude is used to encode would be better.

You can record as many transitions per second as you'd like; AC bias erase signal can be in the hundreds of kHz, for instance, but on playback you will only get a maximum of 24k transitions per second and a Nyquist frequency of 12kHz.

Magnetic recording is a time-discrete sampled system and has much in common with digital systems in that regard. It's not clocked, so the analysis is a bit more complicated in analyzing how the magnetic domains change in the head gap as the tape moves, but there's lots of reading material to be found on how to analyze it.

To your point, the low frequency response means you have to have a certain number of transitions per second; video tape systems get around the requirements of analog video by using actual frequency modulation and recording the FM signal instead of the raw analog video. So if you encode some QAM and the narrowband FM with a carrier frequency of around 6kHz you might find some interesting possibilities.

Thank you for the added information - there's a lot there I hadn't thought about or didn't know.

Sure thing.

 

[cjs]

96 Kbps data rate would be crazy fast ! That would eat floppy disks alive for loading of linear code - such as games. It could fill 256K of memory in just about 30 seconds.

If my calculations are right, floppy diskettes are still faster. Since you mention loading 256K, I'm going to assume early- to mid-16bit era, with the standard IBM PC 5.25" floppy format of nine 512 byte sectors per track, double sided.

This means that in two revolutions, 2/300 second at 300 RPM you'd read 9 KB. From this Shugart SA400 manual I get a track-to-track step time of 20 ms and a settle time of an additional 20 ms. (This is slow because it's a 1970s drive; early '80s drives were considerably faster.) So that gives 1/150+0.02+0.02 = 0.046 s/cylinder, and at 9 k/cylinder that gives 9 * (1/0.0.46) ~= 196 kb/sec.

Of course in those days the main issue was with the machine keeping up with such data transfer rates. This was part of the great convenience of floppies; if you fell behind you could simply pause and wait for the next sector to come around again, whereas with tape if you feel behind you essentially had to abort the entire load.