On Mon, 06 Dec 2010 01:12:05 -0800, Muzaffer Kal <kal@dspia.com>
wrote:
>On Sun, 05 Dec 2010 16:26:58 -0500, Randy Yates <yates@ieee.org>
>wrote:
>
>>On 12/05/2010 02:19 PM, Randy Yates wrote:
>>> On 12/05/2010 11:16 AM, bharat pathak wrote:
>>>> I have a basic doubt. Does the equalizer come before the matched
>>>> filter or after?
>>>>
>>>>
>>>
>>> bharat,
>>>
>>> A matched filter can be thought of as a device that performs a
>>> time-domain correlation of the incoming signal with a known signal
>>> shape. If the channel filtering changes the shape of the filter,
>>> and the equalizer's purpose is to inverse-filter the channel
>>> filtering, what would the answer be?
>>
>>Although now that I think about it, if they're both linear you
>>can swap the order...
>
>What if the equalizer is adaptive ie time-varying? In that case the
>equalizer also tries to synthesize the matched filter too but I am not
>entirely sure if having the known matched filter helps the equalizer
>adapt better.
>--
>Muzaffer Kal
>
>DSPIA INC.
>ASIC/FPGA Design Services
>
>http://www.dspia.com
That's an important consideration. I know of systems that have been
built with just a big, gnarly equalizer intended to sort out the
channel AND the pulse shape in one operation. This means that the EQ
has to be designed with enough degrees of freedom to do both tasks.
If the pulse shape is known, then the EQ can be simplified by running
the known, matched filter prior to it. This lets the EQ be used for
what it is intended, i.e., correcting the channel effects.
So when the pulse shape is known it is often less complex to implement
the (often decimating) matched filter followed by the EQ. Letting
the matched filter do the decimation also helps control the EQ
complexity by letting it be T-spaced rather than some higher sampling
rate.
Strictly speaking the order is not consequential since, as Randy
pointed out, the EQ and matched filter are both linear filters. The
complexity can get pretty crazy if the EQ is first, though, and it
still has to be trained with errors sliced after both filters. This
probably does bad things to the error feedback latency from the slicer
to the EQ.
Eric Jacobsen
Minister of Algorithms
Abineau Communications
http://www.abineau.com
Reply by Randy Yates●December 6, 20102010-12-06
On 12/06/2010 04:13 AM, Frank wrote:
> On Dec 6, 3:52 am, "bharat pathak"<bharat@n_o_s_p_a_m.arithos.com>
> wrote:
>> Thanks Randy.
>>
>> -Bharat
>
>
> It's probably worth noting that the matched filter and the equalizer
> may not operate at the same sampling rate, e.g. root-raised-cosine
> matched filter running at a twice oversampled rate, decimation by 2,
> followed by a symbol-spaced equalizer. So, in the case where the
> matched filter is running at a higher sample rate then the system is
> certainly not linear in that case and the matched filter must come
> first. It is for this reason that I would be inclined just to say that
> the matched filter comes first.
>
> Frank
Hi Frank,
From a processing perspective I see your point, but if the channel
has corrupted the waveshape, what good is a matched filter going to
do?
--
Randy Yates % "My Shangri-la has gone away, fading like
Digital Signal Labs % the Beatles on 'Hey Jude'"
mailto://yates@ieee.org %
http://www.digitalsignallabs.com % 'Shangri-La', *A New World Record*, ELO
Reply by Vladimir Vassilevsky●December 6, 20102010-12-06
bharat pathak wrote:
> I have a basic doubt. Does the equalizer come before the matched
> filter or after?
The order of applying equalizer and matched filter would be unimportant
but the equalizer is usually executed at low sample rate (1 or 2 samples
per symbol). So, filter first, equalizer after.
Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com
Reply by Frank●December 6, 20102010-12-06
On Dec 6, 3:52�am, "bharat pathak" <bharat@n_o_s_p_a_m.arithos.com>
wrote:
> Thanks Randy.
>
> -Bharat
It's probably worth noting that the matched filter and the equalizer
may not operate at the same sampling rate, e.g. root-raised-cosine
matched filter running at a twice oversampled rate, decimation by 2,
followed by a symbol-spaced equalizer. So, in the case where the
matched filter is running at a higher sample rate then the system is
certainly not linear in that case and the matched filter must come
first. It is for this reason that I would be inclined just to say that
the matched filter comes first.
Frank
Reply by Muzaffer Kal●December 6, 20102010-12-06
On Sun, 05 Dec 2010 16:26:58 -0500, Randy Yates <yates@ieee.org>
wrote:
>On 12/05/2010 02:19 PM, Randy Yates wrote:
>> On 12/05/2010 11:16 AM, bharat pathak wrote:
>>> I have a basic doubt. Does the equalizer come before the matched
>>> filter or after?
>>>
>>>
>>
>> bharat,
>>
>> A matched filter can be thought of as a device that performs a
>> time-domain correlation of the incoming signal with a known signal
>> shape. If the channel filtering changes the shape of the filter,
>> and the equalizer's purpose is to inverse-filter the channel
>> filtering, what would the answer be?
>
>Although now that I think about it, if they're both linear you
>can swap the order...
What if the equalizer is adaptive ie time-varying? In that case the
equalizer also tries to synthesize the matched filter too but I am not
entirely sure if having the known matched filter helps the equalizer
adapt better.
--
Muzaffer Kal
DSPIA INC.
ASIC/FPGA Design Services
http://www.dspia.com
Reply by bharat pathak●December 5, 20102010-12-05
Thanks Randy.
-Bharat
Reply by Randy Yates●December 5, 20102010-12-05
On 12/05/2010 02:19 PM, Randy Yates wrote:
> On 12/05/2010 11:16 AM, bharat pathak wrote:
>> I have a basic doubt. Does the equalizer come before the matched
>> filter or after?
>>
>>
>
> bharat,
>
> A matched filter can be thought of as a device that performs a
> time-domain correlation of the incoming signal with a known signal
> shape. If the channel filtering changes the shape of the filter,
> and the equalizer's purpose is to inverse-filter the channel
> filtering, what would the answer be?
Although now that I think about it, if they're both linear you
can swap the order...
--
Randy Yates % "My Shangri-la has gone away, fading like
Digital Signal Labs % the Beatles on 'Hey Jude'"
mailto://yates@ieee.org %
http://www.digitalsignallabs.com % 'Shangri-La', *A New World Record*, ELO
Reply by Randy Yates●December 5, 20102010-12-05
On 12/05/2010 02:19 PM, Randy Yates wrote:
> [...]
> If the channel filtering changes the shape of the filter,
Correction: If the channel filtering changes the shape of the
signal...
--
Randy Yates % "My Shangri-la has gone away, fading like
Digital Signal Labs % the Beatles on 'Hey Jude'"
mailto://yates@ieee.org %
http://www.digitalsignallabs.com % 'Shangri-La', *A New World Record*, ELO
Reply by Randy Yates●December 5, 20102010-12-05
On 12/05/2010 11:16 AM, bharat pathak wrote:
> I have a basic doubt. Does the equalizer come before the matched
> filter or after?
>
>
bharat,
A matched filter can be thought of as a device that performs a
time-domain correlation of the incoming signal with a known signal
shape. If the channel filtering changes the shape of the filter,
and the equalizer's purpose is to inverse-filter the channel
filtering, what would the answer be?
--
Randy Yates % "My Shangri-la has gone away, fading like
Digital Signal Labs % the Beatles on 'Hey Jude'"
mailto://yates@ieee.org %
http://www.digitalsignallabs.com % 'Shangri-La', *A New World Record*, ELO
Reply by bharat pathak●December 5, 20102010-12-05
I have a basic doubt. Does the equalizer come before the matched
filter or after?