---
title: Lab Meeting Minutes 2024/02/02
tags: lab_meeting
---
> Outline
> [TOC]
---
# PERAL Lab Meeting
- 時間：113 年 2 月 2 日 20:00
- 地點：線上
- 線上會議連結 : [Online](https://meet.google.com/zfi-zmnc-qfw)
- 出席者：吳坤熹老師、謝萬霖、劉怡君、田蕙瑜、沈家正、梁宇騰、劉冠伶、繆亭霄、蘇翊荃、陳嘉璐、陳品妤、陳姿綾、陳姿澖

- 會議主題：[LT Network Codes]()
- 主講者: 蘇翊荃
- 主記: 陳品妤

## 會議內容

- 
- m是封包大小，k是封包數量
- 
- decode複雜度小
- 
- d:degree
- 
- 隨機抽兩個packet進行XOR 產生encode packet
- 
- 
- degree隨著decode完成度下降
- 
- Recoding 可以增加 encode packets，但複雜度較高
- 
- LT Network Coding: LT Code + recode
- 選相同degree的
- 
- 
- 
- 
- 為了抽取 d degree的packet -> 使intermediate node 可以reorder(?
- 
- 將特定的source packet 換成另一個source packet
- 
- 互相連通視同一個集合
- 同一集合內可refine
- 
- LTNC 會產生多餘的overhead 在網路中
- 
- LTNC的decode複雜度較RLNC小

---

### 建議&問題
1. P.9 Can it start from y3? [name=yukino]
Ans: [name=] Any node with degree = 1 will be fine.
2. If there is no node with degree = 1, how will the algorithm operate? [name=Selena]
Ans: [name=] It needs to wait until it receives packet with degree 1 so that it can start decoding.
3. P.13 How do you determine the degree? [name=Ellie]
Ans: [name=Ryan]基於機率分布圖使得decode較為容易
4. P.23 LTNC has lower compleixy than RLNC, why does it has higher overhead? [name=Ellie]
Ans: [name=] Lower time complexity, higher space overhead.
5. [name=Louise] P.12 How to select d accordign to the probability distribution?
Ans: [name=] See the figure in P.6.
6. [name=Louise]If it choose the degree with the largest posibility of degree, will it take more time to make the source packet reach the picked degree?
Ans: [name=Ryan] No.
7. [name=Louise] P.6 What is the vertical line around degree=1000?
Ans: [name=Ryan] I don't know.
8. [name=August] If we must have a degree=1 to decode, why the peak value of Robust Soliton Distribute is not degree=1?
Ans: [name=] We only need one packet with degree=1, not every packet. That's why it is not the peak value.
9. [name=August] Should Recoding done in every intermediate node?
Ans: [name=]
10. [name=Ashley]If there's no degree one packets, how to solve it?
Ans: [name=] It can not be handled.
11. [name=] P.12 For the figure in the upper-right corner, you said LTNC is suitable for network traffic with uniform occurrence. What does this mean?
Ans: [name=]The frequency of each degree will be similiar.
12. [name=Ashley]Is it the reason of encoding's frequency large more than decoding's why the overhead of LTNC is higher than RLNC
Ans: [name=Ryan] I don't known.
12. [name=JiaZheng] P.23 Why is the vertical axis called "gossip period"?
Ans: [name=] 互相交換情報，持續收集封包，然後就有辦法 decode，所需的時間.
12. [name=Edger]why does WC(without coding) have gossip?
Ans: [name=] Forwarding
[name=Edger]Yet, LTNC and RLNC also need forwarding, and they need more. Why do they have less gossip?
Ans: [name=] Not sure.
12. [name=Phoebe] How many packets do I have to collect before I can decode?
Ans: [name=] For 5 packets, you need to collect at least 5 packets. Sometimes you need more.
12. [name=Phoebe] Does it have some metric like "code rate"?
Ans: [name=] These are "rateless" codes. It will keep recoding until the destination can decode. In average it will take the same number of packets as the source.
12. [name=Pheobe]P.18 How does it know that it need to change x3 to x7 node to refine?
Ans: [name=] It will try all possibility of the nodes in every set to decide the highest frequency among other packet.
12. P.18 Why do we want the distribution of packets to be uniform? [name=Angela]
Ans: [name=] If you always receive x1, then this does not help you to decode. That's why we want to "refine" and replace it with another packet.
12. This is a paper published in 2010. Does LTNC has newer progress? [name=Angela]
Ans: RaptorQueue is a newer version of LTNC in 2011. [name=Edgar]

12. [name=Edger] According to your presentation, d is smaller than the number of source symbols. How do intermediate nodes determine the d for recoding when it receives encoded packets?
Ans: [name=] Show weather d can be handled by intermedium node.
12. P.20 You only replace packets in the same group. In this example, x6 is not selected [name=Edgar]
Ans: [name=]
12. Why it need to refine since it won't change encoded source packet?[name=Edger]
Ans: [name=]
12. P.15 Does it inspect in the second step?[name=Edger]
Ans: [name=] It will be done in the first step.
12. [name=]Why the symbol x6 is different with other packet?
Ans: [name=] x6 is degree one.
12. [name=] How does the receiver determine the order of decoded packets? (Which is x1? Which is x3?)
Ans: [name=] This information is recorded in the overhead header.
12. [name=Edgar] P.5 What does "belief progagation" mean?
Ans: [name=]
- [Belief propagation_wiki](https://en.wikipedia.org/wiki/Belief_propagation)
12. [name=Solomon]P.15/17 Does it use greedy algorithm?
Ans: [name=]Ordering during preprocess.
12. [name=Solomon] How can an NP-complete be solved
Ans: [name=]

## 會議錄影
https://www.youtube.com/watch?v=zdL9k4CJIko&ab_channel=%E9%99%B3%E5%A7%BF%E7%B6%BE

## 臨時動議

---
散會結束時間： 21:35