Another method of presenting error rate is to consider the errors as the result of adding the data signal to an underlying error signal. The extent of error can then be expressed as the entropy of the error signal, or, in the case of physical signals, as the ratio of the strengths of the two signals — the signal-to-noise ratio — expressed in decibels.
The error rates most frequently specified relate to the following.
A transient (or recoverable) read error occurs during reading and can be recovered by the error recovery procedure prescribed for the storage subsystem (See error recovery). Where the recording format provides sufficient redundancy to allow some error to be recovered on-the-fly, i.e. without re-reading the data, it is necessary to define also the raw error rate, which is the rate that would be perceived if on-the-fly error recovery was not applied.
A permanent (or irrecoverable) read error cannot be recovered by the prescribed error recovery procedure.
A transient (or recoverable) write error occurs during writing and can be recovered by the error recovery procedure prescribed. It is desirable, though not easy, to distinguish two components of this error rate: errors attributable firstly to flaws in the media and secondly to failings of the device (one reason for the difficulty is that these tend to interact).
A permanent (or irrecoverable) write error cannot be recovered by the prescribed procedure. Again it is necessary to distinguish between media flaws and device errors: rather than give a figure for the latter it is usual to regard each occurrence as a fault to be accounted for in the failure rate of the device (See hardware reliability).
An undetected error is an error that is not detected by the storage subsystem, presumably because of some inadequacy in the error check facilities defined by the format or in their implementation, or because of errors occurring outside the ambit of these facilities (See data integrity).