| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
|
Hi Adee, Thanks for the comments, please see my responses inline. Regards, Hossein From: Adee Ran (aran) [mailto:0000147b29386f6c-dmarc-request@xxxxxxxxxxxxxxxxx]
Hossein, thanks for providing your contribution
https://www.ieee802.org/3/dj/public/adhoc/COM/24_1001/shakiba_3dj_COM_02_241001.pdf. I have a few technical comments:
-
On slide 3 “ADC clip level is chosen so that clipping frequency is equal to the target error rate”
o
The clipping noise can reach larger values than the quantization noise, although with lower probability, but it can
lead to undesired effects. In my experience, gain control algorithms typically targets much lower probability of clipping (often using some “soft clipping” level as an additional guard band). [HS] I hope the response to your later comment to parameterize the clipping probability
somewhat addresses this comment. I believe you would agree that implementing “soft clipping” is an overkill here.
o
I don’t see the effect of the clipping noise in the calculations on slide 5. The effect may or may not be negligible
and may depend on the choice of clipping probability. [HS] To simplify, I had only used signal pdf to calculate clipping level. I have now
added noise too. The difference is negligible, but this is a good suggestion.
o
I would suggest making the clipping probability a separate parameter. The initial value can be 2*specBER as currently
used, but we should study the effect of choosing a lower probability and preferably choose something more safe. [HS] As per this suggestion, I added an ADC clipping probability (adc_clip_rate) to the
COM parameters and defaulted it to 2*specBER.
-
On slide 11 “There is a negligible difference in DER_DFE“ – this seems reasonable with ENOB=32, but with ENOB=6
I would expect to see a significant difference. Can you clarify? [HS] Yes, with ENOB=6, as you can see on slide 11, the difference is real and what we
are trying to figure out. With no quantization, represented by a large ENOB (e.g. 32) no difference is expected. The negligible difference is only due to numerical accuracy as explained on the backup slide. With the suggestion by Piers (that I have now implemented)
this numerical accuracy issue is removed altogether, as for the case of no quantization the change is completely bypassed. And some editorial comments, assuming your code is considered for committing:
-
In multiple places you have expressions with the pattern “index
= find(min(vector_expression)==vector_expression)” and then use index(1) to find the (first) location of the minimum. This can be written in
MATLAB using the syntax “[~, index] = min(vector_expression)”.
The latter is somewhat faster to calculate, but more importantly it is easier to read and understand (especially when the vector_expression is long, as it is in your code). It also works with max(), and will result in index being a scalar so you don't have
to take the first element. I would suggest changing the code to make use of this method. [HS] Thanks for the suggestion. I had interchangeably used this syntax before and it
so happened that in this case I used the other method. I have now changed it to this syntax. I was not aware of the difference is execution time.
-
Please change
quantizaion to quantization (in multiple places) [HS] Thanks for the catch, done.
-
On slide 7 you define a few variables initially which are aliases of either existing variables or fields in “param”.
These variables are then used only once. This makes the code longer and harder to track. I suggest using the existing variables and fields instead. [HS] Thanks for the suggestion, done. </Adee> -----Original Message----- Colleagues, The proposed agenda for the 15 Oct COM ad hoc is to review "Commit Request 4p6_5: Quantization Noise Feature". See: https://www.ieee802.org/3/dj/public/adhoc/COM/index.html Note the short duration. Most of the (short) meeting time will be dedicated to code review and Q&A. Use of the electrical email reflector is strongly encouraged.
I want to remind all teleconference meeting participants to review the following documents prior to participation in an IEEE 802.3 meeting teleconference: * IEEE SA patent policy * IEEE SA Copyright Policy * IEEE SA Participation Policy and Code of Ethics All of these policies may be found at
http://ieee802.org/3/policies.html -Kent COM Ad hoc Chair, IEEE P802.3dj Task Force ________________________________________________________________________ To unsubscribe from the STDS-802-3-B400G-ELEC list, click the following link:
https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G-ELEC&A=1 To unsubscribe from the STDS-802-3-B400G-ELEC list, click the following link:
https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G-ELEC&A=1 To unsubscribe from the STDS-802-3-B400G-ELEC list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G-ELEC&A=1 |