Project Kryoflux – Part 3: Recovery in Practice

If you’ve been following along with my previous parts, you can probably already gather that I’m quite enthusiastic about the possibilities offered by the Kryoflux device and doing some “digital archaeology”. But I should probably note (before I get thwacked on the head) that the term Kryoflux is a trademark of the Software Preservation Society and my use of the word is merely in reference to their device.

But it is a cool word – it’s pretty much exactly what I want to do – freeze or preserve (cryogenics) the flux transitions on the disks, and describes everything so elegantly – that’s why I called it “Project Kryoflux”, rather than “Floppy Reading” which sounds much more boring and tedious. Unfortunately, the Kryoflux doesn’t make the job of changing floppies any easier … so the job is still fairly tedious even in 2013 …

But this is where things get interesting – when we actually deal with real life data readbacks. All manner of strange things happen and are worth looking at.


In this little segment, I’ll focus on formats. In my work so far, I’ve focused on IBM/PC floppies which would be the MFM Sector Image format, and the Apple format. Interestingly, possibly for reasons of better PC compatibility, High Density Apple format 3.5″ floppies use the MFM Sector Image format as well as they use the MFM encoding rather than the GCR encoding. The difference is the filesystem on these disks are HFS (rather than FAT12) thus the PC will not recognize the format without some “software help”. The Apple Double Density disks are the ones that really need the Kryoflux help – these are the Apple 400K/800K Sector Image format as they are GCR coded and will not be read by any PC natively.

Choosing the correct sector format results in the read blocks illuminating in green to say they are good and verified as compliant with a selected format. If the format does not match, it may come up grey as “unknown format”, or as yellow with a status of mismatch. For MFM encoded disks, Kryoflux has another feature to detect modified blocks, and these will normally come up as orange with a H in the cell to indicate that additional header data is found and the disk is modified since mastering. For all the disks you write at home, you can consider this the same as good as the disk is modified from the factory formatted state, but for a commercially replicated disk, this is an indication that someone or something may have modified the encoded data.

If you choose to dump to STREAM only, you will only see grey blocks as STREAM does not attempt to interpret the flux transitions and determine the encoding format. So … the message is – choose the right format if you know it. Guess it if you have to.

Apple Double Density Disks

Apple DD disks are special things. Their GCR coding squeezes a varying number of sectors across the disk in several zones to make most efficient use of the media – but it also poses a reading challenge.

I grabbed a professionally mastered GCR 800k format disk and read it first in the Samsung drive (apologies for the screenshots – I did not prepare for this, hence the clutter. Maxtrack was calibrated to 82 tracks to ensure all drives could be switched in and out without recalibration):


It’s a professionally mastered disk, it shouldn’t be like this. Notice how the Scatter plot seems to show some strangeness in the “splitting” of the bands … that seems to be a common thing with all GCR disks and may represent just how difficult it is to read these disks on a fixed speed drive. So then I tried a Panasonic


First of all, it’s important to realize that the results from the two drives are different. Don’t assume a disk that does not read in one drive will not read in another. This is why having a few drives can be plenty helpful! The other thing to realize is that it seems this drive has difficulties with different tracks compared to the Samsung. Lets take out another candidate drive – the Mitsubishi Electric one:


By now, you might have concluded that this disk is bad, and we should give up on it already. But you would be wrong. Each drive has its own head alignment (radial, azimuth, track to track, and side to side), as well as its own unique head width, electronics, sensitivity and AGC settings. In this case, when we stick it into the Sony MPF920-1, we are met with glorious success (as with two Sony MPF920 Z/160’s):


When I stick it into my Teac drive – one of them manages to read it too, but the other doesn’t.


Let me make this clear – all the above drives perform perfectly fine when reading and writing HD MFM formatted floppies. When used to read HD floppies, they work perfectly and are just as good as each other (in general). There are no head alignment issues as such. I found that the only model of drive consistently capable of reading the Apple DD floppies with reliability is the Sony drives.

I went as far as to damage two Panasonic drives by twiddling with their track-zero location, and their radial alignment (by adjusting the worm drive motor rotation) and managed to get the head to go from edge to straddling to the other edge of the tracks but never with consistent success. I also played with the two screws above the head to play with azimuth and side-to-side alignment, which borked everything up even worse. So don’t twiddle the alignment – it doesn’t seem to be an alignment issue.

All Disks

An extension on what I was saying earlier applies to all disks. Just because you can’t read it with one drive doesn’t mean another drive can’t read it – and just because one drive has a problem with one track doesn’t mean another one will.

All of the examples I will show are real disks which have been read in with real drives but have been reproduced from the STREAM files. I didn’t screenshot them initially – as I didn’t have the time to, but it’s worth looking at.

Disk #282 for example, read with Samsung:

Nasty Read

Someone who knows what’s going on will look at the timing and say “this is ugly, maybe your head azimuth is off“. Maybe it is, but since the drive itself read many many disks successfully, it should be within tolerances. But pop it into my trusty Sony drive and …


… same disk, same track, same side and look at how much cleaner the timing diagram looks like? The modified status is a “false positive” which happens when reading is poor because this was a factory fresh formatted disk which sat there for ages. You would think that means the alignment of the disk is right – but I imagine it is borderline.

Let that be a lesson to all of you.

Commonly Encountered Disk Badness

While running the disks through, I find it nice to have the scatter plot open so that any bad things can be spotted right away. It does use some more CPU, but it’s plentiful here, so I just leave it on and watch it carefully when bad sectors arise.

Almost certainly, in my experience, premastered disks and floppies which have been written in one pass tend to be much more reliable and easier to read than those which have been written piecemeal. This is because there are many timing issues which may arise from consumer grade drives and controllers which are less precise when compared to proper duplicating equipment like the Tracer/ST. At one point in time, Trace Mountain Products had a near monopoly on floppy disk duplication equipment – we shall probably get to that in the next installment.

I shall try to show examples of bad disks. The previous one with the timings all going around – that’s an example of how some azimuth misalignment might look like.

First up, the humble “speed variation” – this can happen due to dirty or gummed up bearings, a faulty spindle motor, or a bad disk which doesn’t rotate freely in its shell. It can appear during reads, or writes or both!

speed variation

I have found a certain type of gold-shuttered Verbatim DatalifePlus disk with this issue that is so severe that Kryoflux aborts reading due to the index marks not being consistent. This was solved by orienting the floppy drive vertically – so give that a shot if all else fails.

There is another common issue where there is a erasure spot – a noise segment is read on every revolution at the same spot. This can be media dropout, media contamination interfering with the head contact, accidental erasure by external magnetic field or even possible defective erase head gating. I am not a floppy drive expert, so I can’t say for sure, but any one of these issues could cause this to appear.

spot erasure

Another problem with poorly stored floppies you may encounter is what appears to be sectional shrinkage of the donut. This is where the exposed part is affected by heat and sunlight and curls or shrinks causing a read error at the same area relative to the index mark, normally more severe for the outer tracks. It may appear like this, two noise bands which is the “edges” of the affected area. I actually visually inspected the donut and saw this was the case. Sometimes trying several times can help as the head “flattens out” the media but if it’s really severe, there might be nothing that can be done.

sectional shrinkage

Erasure by magnetic field influence seems to have been witnessed as well – in one case it erased a section of every track with the same phase relationship to the index suggesting a magnet may have passed the shutter area at one time causing one strip of the media to be radially demagnetized.

magnetic field

Or this disk which seems to have either been written with alignment so poor that everything was lost, or more likely, either made with bad media and has suffered entire full-surface bit-rot or written with incorrect coercivity (i.e. HD disk written on a DD drive) and the resultant magnetic field is lost. Interestingly, drives with severe radial misalignment which are reading between or on the shoulders of tracks will show something just like this.


That being said, it isn’t completely lost. You can see hints of the sectors along the track just mixed in with a heap of background noise. A blank unformatted track turns up as noise, like so (it varies slightly depending on the drive):


Other Notes

While recovering a lot of disks, there may come a time when everything seems to be going downhill. Bad disk after bad disk. But all hope may not be lost. Maybe all the drive needs is a head clean. Old disks, especially those stored in poor conditions or heavily used tend to be dirty and carry contaminants into drives. Poor quality media may start to delaminate from the substrate and leave heavy oxide deposits on your heads – you know this if you’ve ever heard a floppy drive squeak. These discs really have one and only one chance to be read (generally), and will possibly even leave a drive out of service once complete.

So this is where a bit of care is needed. I make my own floppy drive cleaning “disk” by borrowing an old shell and stuffing in a low lint fabric soaked in high purity isopropyl alcohol. Make sure the section in the window is well held in and relatively flat – and then command the drives to do some gentle seeking back and forth to scrub the heads. It’s worth a shot.

But sometimes even that won’t work – so it might be time to change drives. You might find that another drive reads consistently, which would indicate that the alignment of the original drive may have been damaged by severe media contamination or just wear and tear. The more recent drives seem to be built relatively flimsily and only designed for infrequent use – so having to “give up” might not be so uncommon. In my course of disk recovery, I had totally consumed the lives of two 3.5″ drives and one 5.25″ drive (one of two). But once they have been read in, they are safe, so despite not wanting to sacrifice drives, this is probably the best cause a drive could sacrifice its life for.

You might also note many of the disks you image may come out with different sized sector images, and some will show 81 tracks used (i.e. 0-80 rather than 0-79 as expected). I will elaborate on this further in the next installment, but it seems some manufacturers of floppy disks formatted their disks in duplication machinery that left a duplicators mark on the 81st track (track 80). This sometimes comes up as a Good+Modified sector, or a Mismatched sector. Don’t worry – that is normal.



AFuji DD 1Snd remember this disk? Well the Kryoflux was able to solve the mystery after all – it’s a single sided disk alright!






After a lot of work and a lot of time, we end up with a stack of neat disk images. These are binary representations of the data which would have been written to the disk. We will also have the STREAM files which represent the flux transitions on the disk, faithfully recording the results of copy protections. But what secrets are awaiting our discovery, and how does one make use of it? Stay tuned for the next installment

About lui_gough

I'm a bit of a nut for electronics, computing, photography, radio, satellite and other technical hobbies. Click for more about me!
This entry was posted in Computing and tagged , , , . Bookmark the permalink.

6 Responses to Project Kryoflux – Part 3: Recovery in Practice

  1. Pingback: Project Kryoflux – Part 4: The Hidden Secrets | Gough's Tech Zone

  2. Pingback: Project Kryoflux – Part 2: Why Bother with it? | Gough's Tech Zone

  3. Caroline Martin says:

    Hi, would you please be able to put up a picture of the “floppy drive cleaning disk” that you made?

    • lui_gough says:

      Dear Caroline,

      I make a more detailed mention of it with two images in Part 6 here:

      If you need more details, let me know, but the idea was to sacrifice an existing floppy and use its liner, high purity isopropyl alcohol, and the calibrate feature of the Kryoflux to gently scrub the heads over the moistened liner which will capture the oxide.

      – Gough

  4. imqqmi says:

    I’ve gone the diy route and implemented reading of direct analogue drive head using an pic mcu as controller, oscilloscope controlled by pc software over ethernet as well as using a stepper motor controller to use microstepping to read between tracks.
    The bad disk you describe with an entire bad disk looks very similar to about 10 disks I encountered and it wasn’t bit rot or misalignment at all. The metal hub in the center of the donut came loose with those disks in my case. I found out while cleaning with my 3D printed cleaning kit I designed. I could feel no resistance while rotating the disk and cleaning the surface. The metal hub spun freely while I was blocking the donut motion.
    I carefully cut the label at the top, cracked two plastic rivets and took out the donut and metal hub. Glued it back together using superglue. I put it back in the case. When reading, about 50% could be read. This still wasn’t caused by butrot but the donut wasn’t perfectly centered. That’s where microstepping comes in handy, reading between the tracks. That way most of the 10 disks were 100% recovered.
    I’ve been able to recover 98.5% of all the 300+ pc and amiga disks I own. Many were heavily fungus infected, stored in damp alternating warm and cold and sawdusty conditions for years. I have a renewed respect for the robustness of floppydisks, despite what others say about it.
    I’m still working on error correction strategies, but so far I’m able to recover up to 2 bytes of data from single burst errors which include missing flux periods, extra flux periods and wrong lengths.


Error: Comment is Missing!