1250MHz ADC

While following the discussion about 8b10b coding, I happened to see a
page on 10GbaseT. It seems that 1250MHz 7 bit ADCs (and 10 bit DACs)
are required. 

Much filtering and echo cancellation is done to the signal, so more
bits are needed than actually get sent down the line. 

I haven't followed ADC technology, but that sounds pretty fast.

-- glen

On 3/29/2013 12:59 AM, glen herrmannsfeldt wrote:
> While following the discussion about 8b10b coding, I happened to see a
> page on 10GbaseT. It seems that 1250MHz 7 bit ADCs (and 10 bit DACs)
> are required.
>
> Much filtering and echo cancellation is done to the signal, so more
> bits are needed than actually get sent down the line.
>
> I haven't followed ADC technology, but that sounds pretty fast.

Single chip GHz class 8-bit ADCs were available since about 15 years 
ago. That made digital oscilloscopes widely available.

VLV

On Fri, 29 Mar 2013 05:59:17 +0000 (UTC), glen herrmannsfeldt
<gah@ugcs.caltech.edu> wrote:

>While following the discussion about 8b10b coding, I happened to see a
>page on 10GbaseT. It seems that 1250MHz 7 bit ADCs (and 10 bit DACs)
>are required. 
>
>Much filtering and echo cancellation is done to the signal, so more
>bits are needed than actually get sent down the line. 
>
>I haven't followed ADC technology, but that sounds pretty fast.
>
>-- glen

I think the challenge these days is keeping the cost and power
consumption down.   They're pretty impressive, IMHO.

Several wireless applications are moving to large bandwidths, so it'll
be interesting to see how that drives the ADC ecosystem.   The
integration levels may prohibit their existing as individual parts,
though.




Eric Jacobsen
Anchor Hill Communications
http://www.anchorhill.com

Eric Jacobsen <eric.jacobsen@ieee.org> wrote:


(snip, I wrote)
>>While following the discussion about 8b10b coding, I happened to see a
>>page on 10GbaseT. It seems that 1250MHz 7 bit ADCs (and 10 bit DACs)
>>are required. 

(snip)
>>I haven't followed ADC technology, but that sounds pretty fast.
 
> I think the challenge these days is keeping the cost and power
> consumption down.   They're pretty impressive, IMHO.
 
> Several wireless applications are moving to large bandwidths, so it'll
> be interesting to see how that drives the ADC ecosystem.   The
> integration levels may prohibit their existing as individual parts,
> though.

I suppose so. It has never been easy to do both analog and digital
on the same chip, though. Well, gigabit (1000baseT) is now fairly
common for consumer (home use) equipment. It might be a little longer
before that is needed at 10GbaseT. 

As to the signal processing, note that both 1000baseT and 10GbaseT
send and receive on all four pairs at the same time. There is a fair
amount of attenuation, so extracting the incoming signal from the
outgoing signal takes a little work. As the impedance match is never
perfect, and often somewhat less, there will at least be a far-end
echo to subtract. Also from any connectors in between. 

So, each end needs four DACs and four ADCs, which at least could be
on the same chip, even if the rest of the digital circuitry is 
separate.

-- glen

On Fri, 29 Mar 2013 19:26:31 +0000 (UTC), glen herrmannsfeldt
<gah@ugcs.caltech.edu> wrote:

>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>
>
>(snip, I wrote)
>>>While following the discussion about 8b10b coding, I happened to see a
>>>page on 10GbaseT. It seems that 1250MHz 7 bit ADCs (and 10 bit DACs)
>>>are required. 
>
>(snip)
>>>I haven't followed ADC technology, but that sounds pretty fast.
> 
>> I think the challenge these days is keeping the cost and power
>> consumption down.   They're pretty impressive, IMHO.
> 
>> Several wireless applications are moving to large bandwidths, so it'll
>> be interesting to see how that drives the ADC ecosystem.   The
>> integration levels may prohibit their existing as individual parts,
>> though.
>
>I suppose so. It has never been easy to do both analog and digital
>on the same chip, though. Well, gigabit (1000baseT) is now fairly
>common for consumer (home use) equipment. It might be a little longer
>before that is needed at 10GbaseT. 
>
>As to the signal processing, note that both 1000baseT and 10GbaseT
>send and receive on all four pairs at the same time. There is a fair
>amount of attenuation, so extracting the incoming signal from the
>outgoing signal takes a little work. As the impedance match is never
>perfect, and often somewhat less, there will at least be a far-end
>echo to subtract. Also from any connectors in between. 
>
>So, each end needs four DACs and four ADCs, which at least could be
>on the same chip, even if the rest of the digital circuitry is 
>separate.
>
>-- glen

I remember when those approaches were being debated in 802.3.   At the
time it was characterised as being like wired MIMO, but I don't know
how close that analogy is.

If history is any guide, cost and size will go down as capabilities go
up, but I suspect ultimately there will be a limit of some sort. 


Eric Jacobsen
Anchor Hill Communications
http://www.anchorhill.com

On Thursday, March 28, 2013 10:59:17 PM UTC-7, glen herrmannsfeldt wrote:
> While following the discussion about 8b10b coding, I happened to see a
> page on 10GbaseT. It seems that 1250MHz 7 bit ADCs (and 10 bit DACs)
> are required. 

The baud rate is 800 MHz so I am not sure why 1250 MHz ADC is needed. All the implementations I know off use a 800 MHz ADC. Maybe someone is doing 2T/3 fractional design ?

> Much filtering and echo cancellation is done to the signal, so more
> bits are needed than actually get sent down the line. 
> 

Actually 10GBT is significantly more complicated than that. One needs echo, NEXT, FEXT and potentially alien NEXT cancellation. In addition to LDPC based FEC, there is THP coding. Basically both sides have local DFE which they converge during training and then the coefficients of the DFE are sent to the other side which uses them to pre-equalize the transmitted data. Of course there is adaptive power back-off which controls the transmit amplifier so that short cables don't transmit too much power in order not to overwhelm the echo canceller.

> I haven't followed ADC technology, but that sounds pretty fast.

Fast ADCs can be done. These days you can buy 3 GHz 8 bit ADCs but the problem is power. In 10GBT people want 4W per port max. This is for all the DSP plus an 800 MHz 8 bit ADC which makes it quite challenging.