Cell Relay Archive

["C. M. Heard" <heard@vvnet.com>]

On Thu, 18 May 2000, novice wrote:
> I'm a graduate student, I'm implementing O.191, O.150, and I have some
> questions.
> 
> 1. Why scrambling is needed for test cells?

This question came up several years ago in an ATM Forum e-mail discussion
group.  According to a contributon from Michel Bonnifait and Christophe
Rabadan (then of France Telecom), the reason of scrambling the test cell
payload is to provide a stress pattern which exercises all bit combinations
in a network element switch fabric.  This provides a better test of switch
fabric transparency than a fixed pattern would.

> 2. Why PRBS used to assess Bit Error Ratio for digital transmission?

PRBS patterns a easy to generate, relatively easy to synchronize, and
provide excellent "randomness" properties -- see answer to next question
for more details.

> 3. What's the effect of the period of PRBS?

The PRBSs (pseudo-random binary sequences defined in O.150 are the
outputs (possibly complemented) of linear feedback shift registers with
maximal period 2^m - 1.  Sequences generated by such shift registers
are also known as m-sequences.  Note that the maximal period of an
m-sequence cannot be 2^m, since a linear generator that starts in
the all-zeroes state stays there forever, by virtue of linearity.

It turns out that m-sequences imitate random sequences in several
interesting ways.  One is that the periodic auto-correlation has only two
values, 2^m - 1 and -1, when one maps the binary sequence to real numbers
via bi-phase modulation (binary 1 -> -1, binary 0 -> +1).  This implies
that the Fourier spectrum is very nearly flat.  Another property is that
all sub-sequences of length m, except m zeroes, occurs exactly once.  A
third property is called the run distribution property:  for each k such
that 0 < k < m-1 there are 2^(m-k-2) runs of 1s and an equal number of
runs of 0s of length k, and there is one run of m-1 1's and one run of
m 0's [note that these last two will be swapped for a complemented
m-sequence, which is how some PRBSs are defined]. For more information see
Solomon Golomb's classic book _Shift Register Sequences_ (Aegean Park Press,
1982).

One of the effects, then, of a longer period is that there are longer
consecutive strings of 1s and 0s.  This is sometimes significant when
testing clock recovery circuits that rely on transitions in NRZ data.

> 4. How can I resynchronize the PRBS at the receiver side?

Assuming that the PRBS is a non-inverted m-sequence just take m consecutive
received bits and use them to seed the generator (i.e., jam-set the shift
register stages with these bits).  To account for the possibility of an
error just check that the next m bits from the generator agree with those
received.  If they do, you are properly synchronized.

If the PRBS is an inverted m-sequence then you need to inverted the received
bits before jam-setting, but the rest of the procedure is the same.

To monitor for sync you need to measure the proportion of errors between
the generator output and the received data.  If memory serves, O.150
suggests that you consider that a loss of sync has occurred if you get
10% errors in one "block" (these groups of consecutive bits, typically
about 1000, with boundaries arbitrarily chosen, into which one divides
the bit stream for the purpose of calculating block error rate).

Mike
--
C. M. Heard
heard@vvnet.com