Přepis řeči - DOWNLINK MULTICELL COOPERATIVE TRANSMISSION WITH IMPERFECT CSI SHARING

0:00:14	the morning um
0:00:16	or this works you know about this work
0:00:18	and do we start the to do you want something make is the coverage you mission we probably
0:00:24	channel state information sharing
0:00:26	a a more efficient
0:00:30	we we to
0:00:31	yeah you the background of rocks and deprive
0:00:34	the model
0:00:36	that's so a scroll to myself a combination strategy for both
0:00:41	so use that information about uh
0:00:44	the
0:00:45	totally
0:00:51	a
0:00:51	you you were also free you you to in the world
0:00:54	is there
0:00:55	the received a a major one that what to be high
0:00:59	so see
0:01:00	uh
0:01:01	and the message that really coverage you transmission and all of them have but they did
0:01:06	my D point commission
0:01:08	is a promising solution to hide we there
0:01:11	yeah we um
0:01:14	to
0:01:14	uh
0:01:15	call it or structure
0:01:17	for a a to much research
0:01:20	relation
0:01:21	oh a multiple
0:01:23	but but station
0:01:24	i you to us
0:01:25	we need to five recordings
0:01:28	um
0:01:28	a a want to be so my people
0:01:31	when you press
0:01:31	incorporating there
0:01:33	we kind of summarise is that
0:01:35	uh
0:01:36	the general procedure predictor of one
0:01:39	click your contribution
0:01:40	you
0:01:41	turns
0:01:42	for follows
0:01:44	for a should station
0:01:46	at is uh
0:01:47	the channel
0:01:48	from to all you for
0:01:51	it is a a a a in right channel reciprocity
0:01:54	and
0:01:55	for work
0:01:55	and four
0:01:56	for that that have to men the node what to the to you need to buy a recording
0:02:01	how to get all all of that all channel from all users
0:02:06	that's it
0:02:07	you of the recorder will all
0:02:09	oh all this issues
0:02:11	and then broadcast
0:02:12	that precoder two
0:02:14	i i to the should by
0:02:17	then um
0:02:18	the the is come from
0:02:20	draw down information
0:02:22	so all you
0:02:24	and we uh uh we do not to use the procedure
0:02:27	yes
0:02:27	so
0:02:28	a a to you need to process just a calls all a precoder
0:02:32	computed exactly
0:02:33	you can use
0:02:37	of the performance gain all considered calm transmission
0:02:40	uh
0:02:41	set on cost or other or at and
0:02:44	but you you know the rules
0:02:46	one to and a channel information sharing
0:02:49	uh
0:02:49	in true
0:02:51	uh and sorry for me a sharing between big
0:02:53	yeah
0:02:54	station
0:02:55	we be perfect
0:02:56	so is that some work
0:02:58	uh
0:02:59	if there is a
0:03:01	a in parallel
0:03:03	oh
0:03:04	in channel state from injury
0:03:06	oh we test
0:03:07	for one uh a uh i U K twelve
0:03:10	and
0:03:11	whichever all
0:03:12	so a car to the region
0:03:15	and every two or three
0:03:16	uh can be don't but he's a with such as from a sharing
0:03:20	to to and the proposed
0:03:22	um
0:03:23	a some to to record
0:03:26	in in a the total available
0:03:28	a a that the
0:03:31	fact
0:03:32	um are you are to be for me
0:03:34	we sharing
0:03:35	yeah i i hundred and see
0:03:37	uh in country deployed
0:03:39	so that runs
0:03:40	the point of it
0:03:42	yeah order of ten
0:03:45	and i do we can find that say
0:03:47	the that
0:03:48	last
0:03:49	try to century use
0:03:51	do you need setting to do this
0:03:53	the channel delay
0:03:54	oh a at of E
0:03:55	to two that
0:03:56	be addition to to to it is to be patient
0:03:59	from for which
0:04:00	maybe you um
0:04:02	to three from and
0:04:03	uh
0:04:05	to the day
0:04:06	information
0:04:07	so i you talk we propose you column
0:04:10	a strategy to anyway says that
0:04:13	the let by
0:04:14	uh recording
0:04:18	a a there uh
0:04:19	and two D comp time with for the to don't not the iterations
0:04:24	and uh uh we can do
0:04:26	C is great you two
0:04:28	uh uh do you
0:04:30	a a job so maybe you uh
0:04:33	for
0:04:33	you the ones in there
0:04:35	know how we can
0:04:36	press that
0:04:38	if you you know what you but it's creation
0:04:41	uh where X i that data you people that i which
0:04:45	sure that be efficient
0:04:47	and do we try to locate role
0:04:51	if you can they should uh i
0:04:53	a but i is the precoding vector all
0:04:56	i
0:04:57	and
0:04:57	the time that i
0:04:59	that and the mean and a variance
0:05:01	yeah
0:05:03	also we can define
0:05:05	uh the the channels
0:05:06	from
0:05:07	okay
0:05:07	to all use this by H L P
0:05:10	later
0:05:13	uh we consider that says that addition
0:05:15	have to the channels
0:05:17	a time segment at all
0:05:19	for
0:05:20	and you have he
0:05:22	we are the channel you name that for every
0:05:25	where uh
0:05:27	have the we need a priority for each all channels that that objects
0:05:30	and that you would be
0:05:31	we not at every week that of the channel
0:05:34	uh but is a question
0:05:36	oh we we K i have to go and
0:05:39	so
0:05:40	estimate
0:05:41	addition
0:05:42	and uh yeah okay K that all the channel estimation error
0:05:45	where is a fees we can also but
0:05:48	in to
0:05:52	here we propose propose a a
0:05:55	you
0:05:56	you drive myself of the G
0:05:59	need to be the issue that
0:06:01	i
0:06:01	uh uh can be
0:06:03	from
0:06:03	for
0:06:04	for a to each station
0:06:06	estimates to
0:06:07	a channel from you to you for us
0:06:09	uh we done in china
0:06:11	do
0:06:12	and the for what uh
0:06:14	and the for channel estimate
0:06:16	i
0:06:18	i get a check from all stations
0:06:22	a given you
0:06:23	rock of the channels
0:06:25	all they the bad things
0:06:27	so i think it's midnight addition kind of no
0:06:31	uh the channel for all permutations
0:06:33	that's a uh
0:06:35	you parents
0:06:35	uh
0:06:36	a while back
0:06:38	is
0:06:40	yeah
0:06:40	then the be efficient and we estimate
0:06:43	a channel from you to all
0:06:45	you all your course
0:06:46	you
0:06:47	i i own precoder with all or or channel state information
0:06:51	and and uh the opposition come from only
0:06:54	one
0:06:55	transition to all you
0:06:58	we some right picture
0:07:00	for uh
0:07:01	the they should process of G
0:07:03	so a
0:07:05	uh
0:07:06	recorders at it's different
0:07:07	efficient at each sign
0:07:09	eventually
0:07:10	i
0:07:11	a to each coefficient just why we it
0:07:14	processing as G
0:07:16	and uh
0:07:18	and all of the coverage of addition
0:07:21	have the same
0:07:22	symbol
0:07:23	you know a a probably "'cause" you don't the
0:07:26	make is the sum rate
0:07:28	you
0:07:29	and uh
0:07:31	at at that the channel and a different stations
0:07:35	a a different by at
0:07:36	uh a should be a at each bit position no uh
0:07:40	uh
0:07:41	include a a a a a a are three possible channels
0:07:44	and and that cost
0:07:46	uh channel
0:07:47	estimates is from
0:07:49	so that they to all you press
0:07:51	and the cost uh
0:07:53	the channel uh
0:07:54	mean
0:07:55	or are they should shed but to you need
0:07:58	and the source of party the
0:08:00	find that the estimated the channel state
0:08:03	mission
0:08:04	uh
0:08:05	uh from a addition to all
0:08:08	you
0:08:09	so S T G by a that a the paper is uh
0:08:13	the problem precoder time problem for the into the remote call team in theory
0:08:18	uh
0:08:19	to you meeting problem
0:08:22	and by
0:08:23	might be show the
0:08:24	suboptimal solution
0:08:26	where
0:08:28	we are one
0:08:29	you make sure of to my
0:08:31	uh
0:08:32	both
0:08:33	you you happens in members
0:08:35	but
0:08:37	which are quite a a a a a a of the sharing of few models
0:08:40	and
0:08:41	a but if you can make first
0:08:43	but also
0:08:44	yeah or that he does not real
0:08:46	a sharing a diffusion variables
0:08:48	reference
0:08:49	a T i one and
0:08:51	uh the sharing of a a a a a a a bit
0:08:54	if you were able
0:08:55	we uh winter the but in a
0:08:58	to reach you
0:09:00	it
0:09:00	and that was hard to the share you models
0:09:03	well come if these
0:09:05	the be all
0:09:06	so
0:09:08	so the basic idea of of of the problem there that uh
0:09:12	uh
0:09:12	each base for
0:09:15	uh
0:09:16	three
0:09:16	from
0:09:17	that's estimate set
0:09:18	uh as the means of the are are recall
0:09:21	and
0:09:22	my own
0:09:23	there
0:09:23	so that
0:09:24	can change is what's to the channel should be able to
0:09:27	uh
0:09:28	to uh to mean estimates are really efficient code and known man
0:09:33	for
0:09:34	efficient
0:09:35	have
0:09:35	to where C
0:09:36	oh well
0:09:37	a local channel
0:09:39	this means that uh
0:09:41	the channel from one tool
0:09:44	uh why is that
0:09:46	for
0:09:47	uh
0:09:48	but you can be don't by
0:09:50	and uh
0:09:52	and the final portion
0:09:53	we we don't by previous
0:09:55	so
0:09:58	or of the channel
0:10:00	channel of each way
0:10:02	uh
0:10:02	okay
0:10:03	we we have that
0:10:04	my
0:10:05	a a game but by outside
0:10:07	the current paper we that
0:10:09	you is that
0:10:10	probably five
0:10:12	uh
0:10:12	to estimate and the yeah that's
0:10:15	a a and channel four
0:10:17	at the time
0:10:18	so it's uh
0:10:20	you you to to mentor the precoder design
0:10:22	at each iteration trends seen the problem
0:10:25	they should know what time
0:10:27	yeah
0:10:28	some uh
0:10:29	that's a
0:10:29	you
0:10:32	we uh we
0:10:34	the which
0:10:34	every a summary and
0:10:36	and
0:10:36	Q
0:10:37	uh yeah i uh
0:10:40	right
0:10:40	a right right
0:10:41	yeah
0:10:43	i
0:10:44	can be approximated by a um
0:10:47	function
0:10:48	here
0:10:49	additional try
0:10:50	can and uh
0:10:51	a by using
0:10:53	uh
0:10:54	change but in into the computer to channel uh channel model
0:10:59	a a that we can formulate the
0:11:01	a vision problem
0:11:03	uh for those uh
0:11:04	we make my
0:11:06	my uh
0:11:07	a it's and the subject
0:11:09	the
0:11:09	proposition one
0:11:11	yeah that okay that
0:11:14	because you image
0:11:15	and
0:11:16	okay
0:11:17	and uh
0:11:19	yeah
0:11:20	the problem is is only a non-convex problem
0:11:23	uh you are to fill so we um
0:11:25	we use the uh the concept of read profile
0:11:28	a you by seven
0:11:30	uh
0:11:32	which you
0:11:33	which he defined as a that we show all of my
0:11:36	which you
0:11:37	it rates
0:11:38	oh we're the
0:11:39	over the rate
0:11:41	so you went
0:11:42	uh
0:11:43	a five we can
0:11:44	my the
0:11:46	i'm rate
0:11:46	okay
0:11:47	uh
0:11:48	mean in all three
0:11:50	and uh
0:11:51	well
0:11:51	for
0:11:52	proposition four
0:11:55	is the um
0:11:57	problem can be re
0:11:59	and you
0:11:59	i rate and and
0:12:01	they four
0:12:02	i i know uh
0:12:03	but it using
0:12:04	to
0:12:05	a
0:12:06	i
0:12:07	and uh
0:12:08	at this not close
0:12:09	a
0:12:10	the
0:12:11	i
0:12:11	one
0:12:12	train
0:12:13	but uh
0:12:14	there a suboptimal solution
0:12:16	the problem can be obtained by in that
0:12:18	estimate that
0:12:19	to pronunciation
0:12:20	a miss it
0:12:21	a given by right
0:12:23	and and one is that
0:12:25	so
0:12:26	no common
0:12:27	a a constraints and that
0:12:29	obtained a to me the programming
0:12:31	but but problem
0:12:34	now we use some rice the yeah
0:12:37	to so called or
0:12:38	and each
0:12:39	for at each base station
0:12:41	you and
0:12:42	file
0:12:43	um
0:12:44	since all that let's
0:12:46	P
0:12:48	the
0:12:48	section true
0:12:49	and a change the optimal side
0:12:52	so
0:12:53	type of rock one that we can
0:12:55	change
0:12:56	and i by
0:12:57	but
0:12:58	competition
0:12:59	a the last uh
0:13:00	and do night vision is so the company
0:13:03	can be used to obtain a
0:13:06	and
0:13:07	P
0:13:07	a one step two to all possible it provides
0:13:11	and and optimal down i
0:13:13	the one thing
0:13:14	well
0:13:15	the model of
0:13:16	some
0:13:18	and
0:13:19	keep important to note that so you you should know
0:13:21	need to
0:13:23	the stuff
0:13:24	two
0:13:24	um
0:13:25	don't
0:13:26	to updating
0:13:27	a coder
0:13:28	to ensure that
0:13:29	a know from um
0:13:30	right
0:13:31	not to with multiple efficient
0:13:33	um
0:13:34	and be
0:13:35	but
0:13:36	uh combine the contract you
0:13:38	uh which can be done by a
0:13:40	right be sure that said
0:13:42	but i mean
0:13:42	i a the right part
0:13:44	he's
0:13:44	i
0:13:46	i
0:13:49	uh
0:13:50	we with it so
0:13:52	a a a real
0:13:53	yeah
0:13:53	true
0:13:54	and uh
0:13:56	to to us in this and that
0:13:58	so
0:13:58	so it ever
0:14:01	uh we didn't read
0:14:02	a time but general
0:14:04	because model
0:14:06	the um
0:14:06	cool
0:14:08	true
0:14:09	are you the pull
0:14:12	a finger also with you
0:14:14	rate
0:14:15	a sure what they are at the
0:14:17	and the uh
0:14:19	such
0:14:20	so T and the whole
0:14:22	if they should present
0:14:24	a
0:14:24	a future
0:14:26	currently for case
0:14:27	you probably
0:14:29	jerry
0:14:30	and you in the case of
0:14:31	both
0:14:32	that is
0:14:33	a i have the same performance
0:14:36	the curly
0:14:37	uh
0:14:38	for
0:14:40	a station
0:14:41	i think that you
0:14:43	curvy
0:14:44	a
0:14:48	that's
0:14:50	that
0:14:51	that
0:14:51	this
0:14:53	G
0:14:54	i
0:14:55	and weights
0:14:56	well
0:14:57	uh and and i think for a dense
0:15:01	okay
0:15:02	conclusion
0:15:04	i
0:15:04	proposed and you only on that
0:15:07	work
0:15:08	uh
0:15:09	G
0:15:12	i
0:15:13	P
0:15:14	uh
0:15:15	a to to
0:15:17	right
0:15:18	that because the U
0:15:19	if and you for providing a
0:15:22	or
0:15:23	the
0:15:23	uh
0:15:24	position
0:15:26	for that G
0:15:28	a a it had a very high complexity
0:15:30	male
0:15:32	you do sign
0:15:33	black
0:15:34	and and uh
0:15:35	that
0:15:36	efficient use it all out
0:15:38	i
0:15:38	should
0:15:41	you
0:16:11	uh
0:16:19	the
0:16:19	shen
0:16:20	and so this uh
0:16:34	uh
0:16:35	yes actually in the
0:16:37	you know that
0:16:39	we you don't you
0:16:41	the case of can
0:16:43	what
0:16:43	only five uh
0:16:45	one uh
0:16:47	uh
0:16:48	maybe more
0:16:50	yeah
0:16:51	B

DOWNLINK MULTICELL COOPERATIVE TRANSMISSION WITH IMPERFECT CSI SHARING

Distributed and Cooperative Processing

Přednášející: Shengqian Han, Autoři: Shengqian Han, Chenyang Yang, Beihang University, China