UOV parameters

823 views
Skip to first unread message

Bo-Yin Yang

unread,
Jun 19, 2022, 4:01:52 AM6/19/22
to pqc-forum
Dear PQC-list:

In view of the recent Beullens attack on Rainbow and the new NIST upcoming call stressing longstanding digital signature schemes, it is surely important to select some parameter sets for Unbalanced Oil and Vinegar, the most venerable non-lattice extant PQ digital signature scheme.

Our team has examined the design space and come up with the following parameter sets and will implement them for the public good.  The numbers are (where q is the field size, n is #oil+#vinegar, and m=#oil) for NIST security levels 1, 3, and 5.

q: 16,   n: 160, m: 64, (SL1)
q: 256, n: 112, m: 44, (SL1)
q: 256, n: 184, m: 72, (SL3)
q: 256, n: 244, m: 96, (SL5)

best wishes,
Ward Beullens, Ming-Shing Chen, Jintai Ding, Matthias J. Kannwischer, Jacques Patarin, Albrecht Petzoldt, Dieter Schmidt, Chengdong Tao, and Bo-Yin Yang

Daniel Apon

unread,
Jun 19, 2022, 5:40:53 PM6/19/22
to Bo-Yin Yang, pqc-forum
Thanks, guys.

Have you benchmarked these new parameter sets for clock cycles of KG/Sign/Verify?

--Daniel

--
You received this message because you are subscribed to the Google Groups "pqc-forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pqc-forum+...@list.nist.gov.
To view this discussion on the web visit https://groups.google.com/a/list.nist.gov/d/msgid/pqc-forum/16e85f78-a37a-4213-81cb-f2dd48d85e02n%40list.nist.gov.

Bo Lin

unread,
Jun 20, 2022, 2:39:10 AM6/20/22
to Bo-Yin Yang, pqc-forum
Hi,

Thanks for the update.

Does this mean that the key size of UOV is as follows?
q n m key size (bits) key size (bytes)
 (q*m*n^2)/2 (key size (bits)) / 8
16 160 64 5,242,880 655,360
256 122 44 30,232,576 3,779,072
256 184 72 122,093,568 15,261,696
256 244 96 287,834,112 35,979,264

Regards,

Bo

--

Matthias Kannwischer

unread,
Jun 20, 2022, 3:07:22 AM6/20/22
to Bo Lin, Bo-Yin Yang, pqc-forum
Dear Bo, 

No, the public-key size of plain UOV is log_256(q)*m*n*(n+1)/2 bytes.

So, you have public keys of  
412160 bytes
278432 bytes
1225440 bytes
2869440 bytes

Cheers, 
Matthias 

Bo Lin

unread,
Jun 20, 2022, 4:24:55 AM6/20/22
to Matthias Kannwischer, Bo-Yin Yang, pqc-forum
Thanks Matthias.
Yes, indeed, I missed the log operation and misplaced the m and n in the Spreadsheet. The (n+1) item was not known to me though. The following table is a re-calculation according to your correction:
parameters
key size (bits) key size (bytes)
q log(q) n m  (log(q)*m*n*(n+1))/2
(key size (bits)) / 8
16 4 160 64 3,297,280 412,160
256 8 122 44 2,641,056 330,132
256 8 184 72 9,803,520 1,225,440
256 8 244 96 22,955,520 2,869,440

Matthias Kannwischer

unread,
Jun 20, 2022, 4:38:08 AM6/20/22
to Bo Lin, Bo-Yin Yang, pqc-forum
Dear Bo, 

n should be 112 for the second parameter set. 

Cheers, 
Matthias

wa...@beullens.com

unread,
Jun 20, 2022, 5:01:45 AM6/20/22
to Matthias Kannwischer, Bo Lin, Bo-Yin Yang, pqc-forum

Dear all,

Note that a large part of a UOV public key is uniformly random and can be expanded from a short seed. (see https://eprint.iacr.org/2010/424 for the details)

So, the public key size is really only log(q)m^2(m+1)/2 bits + a short seed.

This reduces the public key size from 412.2KB to 66.6 KB for the first parameter set.

All the best,
Ward

 

Bo Lin

unread,
Jun 20, 2022, 6:08:06 AM6/20/22
to Matthias Kannwischer, Bo-Yin Yang, pqc-forum
Hi, Matthias,

Thanks for the correction. It is ratified as

parameters key size (bits) key size (bytes)
q log(q) n m  (log(q)*m*n*(n+1))/2 (key size (bits)) / 8
16 4 160 64 3,297,280 412,160
256 8 112 44 2,227,456 278,432
256 8 184 72 9,803,520 1,225,440
256 8 244 96 22,955,520 2,869,440

Matthias Kannwischer

unread,
Jun 21, 2022, 4:26:16 AM6/21/22
to Daniel Apon, Bo-Yin Yang, pqc-forum
Dear Daniel,

thanks for your question about performance benchmarks.

We are currently finalizing AVX2, SSSE3, Arm Neon, and Arm Cortex-M4 implementations of the proposed UOV parameter sets and we will be posting the benchmarking results on the forum soon.

Best regards,
The Team

andy yi

unread,
Jun 22, 2022, 8:45:16 PM6/22/22
to pqc-forum, Matthias Kannwischer, mos...@gmail.com, pqc-forum, Daniel Apon
Does anyone know what the Rainbow algorithm is about submitting the updated parameters?

Jintai Ding

unread,
Jun 22, 2022, 8:50:36 PM6/22/22
to andy yi, Daniel Apon, Matthias Kannwischer, mos...@gmail.com, pqc-forum
Yes. Soon, we are doing final check our numbers. 

--
Jintai Ding, PhD
Professor of Mathematics
Tsinghua University 
Reply all
Reply to author
Forward
0 new messages