# Huawei Marathon

### Open Positions

We have interesting positions for you

### Results

You can see the results here

If you have any feedback, suggestions, doubts on evaluation of your solution, cooperation interest according to the comment below write to: Andrey.Vorobyev (at) huawei (dot) com.

Comments to tasks from Andrey Vorobyev:

1. First task is mainly related to DPD model selection. It is open problem now.
Extended memory polynomial is not best model. With some Volterra terms, EVM can achieve -50 dBc for your signals.

2. For model adaptation we can use Least Squares algorithms (“pinv”, “cholesky”, “svd”) from Matlab or some libraries. Another way is LMS algorithms, mainly implemented in FPGA/ASIC. Final result is same but for LMS we need more time for convergence.

3. Second task has two directions. One way is to use some more complex models, like “Wiener Model” of “Neural Networks”. In that case number if model coefficients can be reduced several times. Another way is apply Orthogonal Matching Pursuit (OMP) algorithms (or others) for “Sparcity-constrained Sparce Coding Problem”, which is our problem with fixed number of coefficients;

4. Last task is “Discrete Optimization” task. Usually we can round float coefficients into integers with 16 bits without EVM degradation. For less than 8 bits “Discrete Optimization” is required. This is also open problem.
For my opinion such tasks need more time (several weeks). It is better do it offline. In that case results will be much better.
If this research area is interesting, we are open for cooperation. One direction is search of new models (“Neural Networks” or others). Second is design special soft for universal DPD models simulation.

### Prizes

### Problem Description

Presenation with data

Helpful article

More signals

Send the results to: Andrey.Vorobyev (at) huawei (dot) com, Andrey.Golomazov (at) huawei (dot) com.

You should send the following files:

- The applications for each criteria you solved (written in C++ or Matlab).
- The files of adaptation step for criterias you solved ("coefficients1.dat", "coefficients2.dat", "coefficients3.dat").
- The modeled signals for each criteria you solved.

### Rules

- The team must be registered via the link until February, 03, 12:00.
- The teams participate in the contest from the start of the camp to February, 8, 02:59.
- The team must consist of no more than 3 participants.

### Questions & Answers

Questions | Answers |
---|---|

1) Do you allow to participate in the competition additionally to students to others attendees (e.g., coaches who attend the camp) ?
2) Can team consist from students from different universities? 3) Is any feedback available when sending the solutions: at least check the correctness of the format? 4) Is it true that the EVM is negative for an adequate response? 5) How many prizes are allocated for one team won a prize? |
1) If there are 3 former ACM finals participants build up a team, yes, we allowed.
2) Yes, it is allowed, if it is more interesting give more chance to solve the task, but every team consists from 3 students. 3) We are ready to answer any questions online by email. 4) EVM < 0; 5) We prepared prizes for the first 3 teams (9 prizes) and the next 8 teams (24 prizes), means totally 33 winners (11 teams). |