 |
|
|
|
 |
|
||
 |
|
||
 |
 |
|
|
|
|
|||||
|
|
|
||||
|
|
 |
||||
|
|||||
|
|
Too bad it's not sexier
than this, but . . . *The channel response is also called the impulse response. This is the head's readback response to the magnetic pattern written on the disk when a pulse of write current is sent to the head. The readback response is a di-pulse. We could also ask "What's the best car?" If there is a best car, why aren't we all driving it? Have you tried to put two child car seats in a Ferrari? There are a lot of different cars because some people need space, some need near zero maintenance (ask Chuck about his FIAT!), others need safety at any cost, some need basic transportation at the lowest cost, etc. Read channels vary in cost, availability, flexibility, adaptability, support, error monitoring, speed, power consumption, execution of design, quality of manufacturing process, variability, pin count, ease-of-use, support tools, etc. The point is this: The best combination of tradeoffs for one manufacturer may not be best for another. That's why one manufacturer isn't making all the read channels! What about comparing two channels for the same application? Well in this case, here's the plot that really tells the tale. Imagine a two dimensional plot. The vertical axis is the SNR required to achieve a certain bit error rate (BER), usually something like 10 -5. Lower is better on this axis. The horizontal axis of this plot is the User Density, measured in PW50/(User bit cell time). This is a unit-less quantity. You can think of it as a measure of inter-symbol interference (ISI). When the channels are compared on this plot, the one that has the lowest SNR requirement over the largest range of User Densities is the winner. Then, of course, comes the issues of cost, power, availability, flexibility, etc. A good reference that uses this plot to compare channel simulations is: Ke Han and Richard Spencer, "Comparison of Different Detection Techniques for Digital Magnetic Recording Channels," IEEE Transactions on Magnetics, Vol. 31, No. 2, March 1995, pp. 1128-1133. Based on their simulations, you can determine the BEST CHANNEL! Well, actually you can determine the best simulated channel under the conditions that were simulated. When you dig deeper you will discover two universal truths of read channels: Universal Read Channel Truth #1. ANY detector can be shown to be the best detector. What? It depends on the conditions you are modeling. For example, if you create a detector that is specifically designed to detect your signal in the presence of a certain non-linear disturbance (e.g. non-linear transition shift (NLTS)), then this detector will be the winner in your study. If you change the noise/distortion conditions to be heavily weighted to, let's say inter-track interference, another detector (one designed for just such a problem) will be the winner. Universal Read Channel Truth #2 Hardware detectors don't match the simulation. Let's say you have completely modeled the noise/distortion conditions that are most important to you. You simulate a wide range of detectors and ask two different semiconductor manufactures to make read channels for you using the same detector architecture. They'll perform the same, right? Not by a long shot. Two big issues right off include how cleverly each manufacturer turns your architecture into a transistor-level design and how good their respective manufacturing processes are. There can be as much as 1dB or more difference between the manufacturers' finished products -- Especially after you take into account that one may have more robust gain and timing loops than the other. This is why hard disk drive manufactures are not too likely to change detector architecture if simulations show a few tenths of a dB improvement is be possible. They are more likely to tweak the current generation of their hardware to have a more realistic chance of achieving a couple of tenths of a dB improvement. When you've got a half dB of SNR improvement to show with simulation, then you might get some attention. In closing this section, we mention another popular comparison test: the bath tub curve. The vertical axis is BER, the horizontal axis is reader position relative to track center. The farther off-track the head can be pushed and still maintain an acceptable error rate, the better the channel. This is also a good way to compare different heads with one channel. The tests are performed at a fixed SNR and User Density. Maxentropic Rate of RLL Codes See DDT FAQs. Includes a table of maxentropic rates for various code constraints. What do PRML Signals Look Like? LeCroy, the oscilloscope manufacturer, has a variety of waveforms available at its disk drive analyzer site. This page takes a while to load. For a college-level treatment of PRML: Jan W.M. Bergmans, Digital Baseband Transmission and Recording, Kluwer Academic Publishers, 1996, ISBN 0-7923-9775-4. The classic paper: This paper detailed, for the first time, a complete working PRML channel. It is from IBM Research in Zurich. R.D.Cideciyan, et al., "A PRML System for Digital Magnetic Recording," IEEE Journal on Selected Areas in Communications, Vol. 10, No. 1, January 1992, pp. 38 - 56. Please let us know if you have a question or comment. Please contact us for a free initial consultation. e-mail:
connect@ChannelScience.com |
|
||
|
|||||
|
|||||
|
|
||||
|
 |
||||
|
|||||