Re: [8023-10GEPON] FW: Bridge details
Dear Frank and Glen,
Based on Frank's explanation, I think the number of correlators depend
on the processing time of correlator. The process consists of 66XOR
and counting the bits that don't match.
processing time number of correlator
1bit (1 clock of 10.3125GHz) 1
16bits (1 clock of 645MHz) 16
66bits (1 clock of 156.25MHz) 66
It is difficult to judge whether the hardware size of the correlators
is small or large. Compared with the hardware size of FEC decorder,
that would be small (maybe less than 1/10), even if 66 correlators are
implemented.
But if the shorter delimeter satisfies requirement, such as the
probability of lost burst and false lock, the shorter delimeter is fine
from the viewpoint of hardware size. when I saw the G984.3 (GPON) that
recommends 20 bits delimeter as burst mode overhead. So I thought why
10G-EPON needs 64 bits delimeter.
Best Regards,
Fumio Daido
Glen Kramer wrote:
> Frank and all,
>
> I would add that this approach requires reading between 66 and 131 bits
> (assuming that number of errors does not exceed the threshold, otherwise we
> would continue reading). Of course, it can also be further optimized s.t.
> once we found 55 matching bits, we may stop, since the remaining bits, even
> if all mismatched, are below the error threshold.
>
> I think Daido-san's question was about my initial proposal to lock on block
> boundary first, and then do block-by-block comparison to find the delimiter.
>
> To lock on block boundary, we use the same technique that Frank described.
> Read bit stream and compare to known sync pattern. Sync pattern is selected
> such that at every shift position, except the matching position, it has at
> least 30 bits different (take a give few for possible errors). Once we found
> a position with small number of mismatches (say, less then 15), we declare
> lock. This is a very robust scheme, as it tolerates a large number of
> errors per block.
>
> Once we locked on block boundary (stage 1), continue comparison block by
> block (without bit shifting), looking for the delimiter value (stage 2).
>
> Glen
>
>
>
>
>> -----Original Message-----
>> From: EffenbergerFrank 73695 [mailto:feffenberger@HUAWEI.COM]
>> Sent: Thursday, December 14, 2006 10:22 PM
>> To: STDS-802-3-10GEPON@listserv.ieee.org
>> Subject: Re: [8023-10GEPON] FW: Bridge details
>>
>> Dear all,
>> On the number of correlators needed to do a 66 bit correlation...
>> I can answer this question like this:
>> If the correlator operates serially, then you need 1 correlator.
>> That is, if this correlator circuit checks each bit alignment in one bit
>> time, that you need only one.
>>
>> If the correlator is working with a 16 bit wide data bus, then you need 16
>> correlator circuits.
>>
>> What's involved in a correlator circuit, you might ask? Fundamentally, 66
>> XOR gates, plus some adding logic to count the bits that don't match.
>> This is somewhat simplified in that once the count goes over 11 bits, we
>> don't care (12 or more mismatching bits is declared a non-lock). I can't
>> imagine that this goes over 200 hundred gates.
>>
>> Sincerely,
>> Frank Effenberger
>>
>>
>>> From: Fumio Daido [daido-fumio@sei.co.jp]
>>> Sent: Thursday, December 14, 2006 12:11 PM
>>> To: glen.kramer@teknovus.com
>>> Cc: 'Frank Effenberger'; Jeff Mandin; kj-tanaka@kddilabs.jp;
>>> marek.hajduczenia@siemens.com; 'Ryan Hirth'; 'Frank Chang';
>>> piers.dawe@AVAGOTECH.COM; 'Shoichiro Seno'; k-shiraishi@bq.jp.nec.com
>>> Subject: Re: Bridge details
>>>
>>> Glen,
>>>
>>> I reviewed proposals after FEC adhoc tel conference. I will show
>>> you my concern as follows.
>>>
>>> Could you estimate how many 66 bits correlator to detect 66B
>>> boundary are needed for instant lock of 66B block? I think pararel
>>> implementaion is needed for quick lock. I am concerned about size
>>> of hardware for detecting 66B block boundary.
>>> If 64 bits delimiter (exclude 2bit sync bits) is large, 64 bits
>>> delimeter can be divided into 32 or 16 bits delimeter.
>>>
>>> Regards,
>>> Fumio Daido
>>>
>>>
>