Image Map Image Map
Page 1 of 3 123 LastLast
Results 1 to 10 of 21

Thread: MFM and Double Density

  1. #1
    Join Date
    Feb 2006
    Location
    in the basement
    Posts
    764

    Default MFM and Double Density

    Hello Folks,

    I guess I don't quite understand the story about the MFM data recording;

    1- If there is no clock flux-transitions for consecutive 0s, how are 0s
    distinguished from one another?

    2- What is the relationship between having less clock flux-transitions in
    MFM and recording more data in a given space?


    ziloo

  2. #2
    Join Date
    Jan 2007
    Location
    Pacific Northwest, USA
    Posts
    25,233
    Blog Entries
    20

    Default

    Look at it from a frequency spectrum standpoint.

    In FM, you have, for each, bit cell, a clock bit and a data bit. So, for a zero, you have a clock bit and no data bit. For a one, you have a clock bit and a data bit. So, to encode 0110 with clock bits we have 10 11 11 10. You can see that there are only two frequencies invovled, f where bit cell transitions are adjacent (11111111....) and f/2 where they are separated by zeroes (1010101010). That's what you'd see on a spectrum analyzer.

    It's not so much a matter clock bits, but maintaining synchronization within the data stream. That is, we have to have a flux transition every so often to keep our decoding on track.

    So how do we double the data rate? Well, FM is essentially a history-free encoding method--you don't need to know what follows or precedes an FM encoded cell--there are only two possibilities 10 and 11. (clock plus no transition or clock plus transition).

    What if we put some history into the encoding, such that a zero is encoded as 10 if preceded by a zero, and 00 if preceded by a one; a one is always encoded as 01. Note that 01100 then gets encoded as 1001010010. Note that you can have 101010101010 patterns, or f/2 at worst, since "1111..." is not a possible encoding.

    Thus we can double the encoding frequency and still stay within the bandpass of the drive electronics. The price paid is that encoding and decoding get slightly more involved because MFM has to aware of its history.

    A little simplified, but you get the idea.

  3. #3
    Join Date
    Feb 2006
    Location
    in the basement
    Posts
    764

    Default

    Quote Originally Posted by Chuck(G) View Post
    ....Thus we can double the encoding frequency and still stay within the bandpass of the drive electronics.....
    Here the word frequency is related to how many more bits we manage to transfer
    at the same Baud Rate...........is that right? We haven't changed the actual time cycles...


    Quote Originally Posted by Chuck(G) View Post
    .......It's not so much a matter clock bits, but maintaining synchronization within the data stream.
    That is, we have to have a flux transition every so often to keep our decoding on track.
    So, the hardware has a synchronization section that takes a kicking
    once in a while and keeps on ticking........is that right?


    ziloo

  4. #4

    Default

    Quote Originally Posted by ziloo View Post
    Here the word frequency is related to how many more bits we manage to transfer
    at the same Baud Rate...........is that right? We haven't changed the actual time cycles...
    It's not about encoding more bits in the same 'bit time', it's about changing the specific data that gets written to avoid having 1's close together. If there are never consecutive 1's in the bitstream you can pack them together closer. And because the drive is spinning at a fixed speed bits that are closer together can be read/written in less time. So not only does MFM pack more data in the same space, it makes data transfer faster because you read/write double the data in the same amount of time.

    My recollection about magnetic recording is that the disk actually records data not as 1's and 0's but as "flux transitions" -- changes in magnetism from 'N' to 'S'. A 1 is written as a change, while a 0 is written as no change. The upper limit on how much data you can store on a disk is determined by how quickly you can write 1's and still be able to read the data back reliably. This translates into a minimum amount of time that must pass after writing a 1, before you can write another 1.

    Since MFM guarantees that you never have consecutive 1's, it makes a piece of media that was rated for say, 1000 flux-transitions-per-inch able to store 1000 bits using MFM. With FM that same media could only hold 500 bits because 500 '1' bits turns into 1000 '1' bits on the media (500 '1' clock bits plus 500 '1' data bits).
    Last edited by kgober; December 5th, 2017 at 05:40 AM.

  5. #5

    Default

    FM had a constant clock. Data could be there or not there depending of if it were a one or zero.
    MFM only had clocks if there wasn't enough data changes or if it was making special marks, like for the
    beginning of a sector.
    The pll for the data separation had to run by using either data or clock to sync on. The reason one needed
    to only have a specific maximum span without seeing either a data pulse or a clock pulse was so that the
    pll wouldn't loose phase position of either the clock or or data.
    Depending on the data being recorded, the MFM could have more or less transition per inch than FM.
    Maintaining phase timing was critical for MFM because clock pulses were only used to fill in for places
    where data was static.
    On the media there was what was called emphasis and de-emphasis. This was because the bits were written
    as N orS as was mentioned. The read heads only see the edges between the N/S, because for noise immunity
    of digital data, it needs to saturate the amplifiers. The problem is that the signal before and the signal after
    the current signal are magnetic fields that can make the transition between N/S shift earlier or latter.
    Two N's in a row will gang up and make the next N>S transition read later in time. The same two Ns can make the
    leading S>N appear earlier.

  6. #6
    Join Date
    Jan 2007
    Location
    Pacific Northwest, USA
    Posts
    25,233
    Blog Entries
    20

    Default

    Quote Originally Posted by ziloo View Post
    Here the word frequency is related to how many more bits we manage to transfer
    at the same Baud Rate...........is that right? We haven't changed the actual time cycles...
    No, strictly speaking, frequency is only loosely tied to bit rate. Consider more aggressive encoding techniques, of which MFM is only one. (2,7)RLL (used extensively on hard disks) has the same frequency bandpass as MFM, but encodes 50% more. We're talking about what a spectrum analyzer would see. Remember that, in a signal processing sense, disk drives are still analog devices.

    Compare with modem technology and the misuse of the word "baud". A 14,400 bit/sec modem runs at 2400 baud, that is, it encodes 2400 symbols per second.

    So, the hardware has a synchronization section that takes a kicking
    once in a while and keeps on ticking........is that right?
    Pretty much--it needs to be reminded of where the bit boundaries (simply speaking) are, since, in spite of our best efforts, mechanical devices that spin do not spin with perfect accuracy. There are both long-term and instantaneous (ISV) speed variations and the electronics needs to be able to track this. It can be done using analog or digital methods.

    One thing to keep in mind that I mentioned was "transitions". Disks use saturation-recording, unlike old audio and video tape recorders. A domain is viewed as either "north" or "south" without (hopefully) regard to degree of saturation. Every time the magnetization reverses, you get a pulse induced in the read heads. Hard disks use a considerably more advanced system of reading (e.g. GMR, PRML), but the idea's the same.

    Look at the input of any floppy drive--somewhere along the write channel, you'll see a flip-flop that toggles every time a one bit is detected. This changes the magnetization in the write heads. Note that the read circuitry is oriented toward detecting pulses, not levels, so a reversal, regardless of direction, produces an output pulse.

    As Dwight mentioned, there are real-world considerations to be taken into account as transitions are pushed closer together, so techniques such as pre-compensation that shifts the transition writing time a bit early or late, depending on what's been written and what will be written. Another, eariler technique is to reduce the writing current as bits get packed closer together.
    Last edited by Chuck(G); December 5th, 2017 at 08:10 AM.

  7. #7
    Join Date
    Feb 2006
    Location
    in the basement
    Posts
    764

    Default

    Is it safe to say that, theoretically speaking, FM encoding signifies
    the minimum safe distance that pulses can be placed next to each other?


    ziloo

  8. #8
    Join Date
    Jan 2007
    Location
    Pacific Northwest, USA
    Posts
    25,233
    Blog Entries
    20

    Default

    In my opinion, no, it's not safe to make that generalization, as it conflates the issue of encoding with what amounts to analog read channel design. For example, one can encode information on a floppy disk using PPM, with the same restrictions on bandwidth imposed by read channel design as in FM or MFM.

    For example, a 3.5" 1.44M media drive adjusts its read channel characteristics according to the media type inserted into it, a primarily analog process.

    There's another issue at play here as well. Advanced encoding schemes such as (2,7) RLL are more sensitive to timing differences, which is why they're used on hard disks (fast spinning with lots of momentum, so relatively constant short-term speed regulation), but not on floppy drives (relatively poor short-term speed regulation). The problem is characterized as "jitter", which is also why hard drives with plated media do better with RLL than ones with coarser bits of iron oxide as the medium.

    Remember that floppy drives are "dumb" devices, with most of the "intelligence" directed at optimizing head positioning.
    Last edited by Chuck(G); December 5th, 2017 at 11:46 AM.

  9. #9
    Join Date
    Feb 2006
    Location
    in the basement
    Posts
    764

    Default

    Alrighty then....let me rephrase my statement; comparing two modes of
    modulation, namely FM and MFM, we can say that on the same media
    the shortest safe distance between two pulses (flux-transitions) is that
    of an FM........Yes....NO...?

    ziloo

  10. #10
    Join Date
    Jan 2007
    Location
    Pacific Northwest, USA
    Posts
    25,233
    Blog Entries
    20

    Default

    Again, it's not tied to any specific encoding scheme. FM is relatively inefficient in terms of encoding, where MFM is more efficient. In both cases, the high-frequency limit is the same; but MFM manages to encode twice as much data as FM in the same space.

Bookmarks

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •