Spaces

Hi 5G NR Experts,
The CORESET has 12 search spaces, which implicitly we can say for 1 CORESET we have have 12 searchspace CORESETs.
So my confusion is: are all the search space (12 search spaces) with the same size of the same CORESET that they are belong to?
Can’t be search space with different sizes for same CORESET?
Each CORESET has maximum 12 search space at grid? In other words 12 CORESETs on grid for same CORESET.
In addition, each search space that belongs to same CORESET can have different search sizes?

Thanks alot

Where did 12 come from? Max search spaces per BWP is 10.
Search spaces are not necessarily non overlapping parts of the coreset. They could just have different ALs

2 Likes

Ah yes sorry wasn’t noticed.

I didn’t understand your point about search spaces?
Do you mean that at each search space has its own ALs? for instance for
searchspace1 <-----> has ALs 2
searchspace2 <-----> has ALs 8



etc.

Yes? So can I imagine that each searchspace is a group of RBs that’re accumulated with one specific AGLs for that searchspace? Means that each searchspace has its own ALs.

Another question, from where the UE knows the size of each CORESET?
In SIB1 from CORESET0, yes?
Appreciated

Any help please?

Hi, @Ryan

Let`s divide and conquer.

  1. “Another question, from where the UE knows the size of each CORESET?
    In SIB1 from CORESET0, yes?”

Almost! MIB carries pdcch-ConfigSIB1 which is the first thing the UE needs to decode and obtain the controlResourceSetZero and searchSpaceZero

From 38.331: PDCCH-ConfigSIB1: The IE PDCCH-ConfigSIB1 is used to configure CORESET#0 and search space#0.

More information here: 5G | ShareTechnote

  1. “Yes? So can I imagine that each searchspace is a group of RBs that’re accumulated with one specific AGLs for that searchspace? Means that each searchspace has its own ALs.”

Almost! Search Spaces are a group of resources in the frequency and time domain which are used with for the porpoise of decoding common and UE-specific information.

Again, more information here: 5G | ShareTechnote but a key highlight is:
“PDCCH Search Space refers to the area in the downlink resource grid where PDCCH may be carried. UE perform blind decoding throughout these search space trying to find PDCCH data (i.e, DCI). At high level view, NR Search Space concept is similar to LTE Search Space, but there are many differences in terms of the details.”

  1. “searchspace1 <-----> has ALs 2
    searchspace2 <-----> has ALs 8”

No! For each search space multiple AggLvls are provided. A set of candidates is provided to the UE via IE nrofCandidates.

1 Like
  1. The CORESET has 12 search spaces, which implicitly we can say for 1 CORESET we have have 12 searchspace CORESETs.

Some important info here from NR in Bullets: a maximum of 3 CORESETs can be configured for an active BWP, even the IE controlResourceSetld having a maximum of 12, these other IDs must be used for different BWPs. The reference here is R15, if R16 introduced changes it is not covered here.

It is not clear to me the maximum amount of SearchSpaces per CORESET. According to ShareTechnote it depends on the SearchSpace type (common or UE-specific), take a look here: 5G | ShareTechnote but ir varies between 4 or 10.

  1. “So my confusion is: are all the search space (12 search spaces) with the same size of the same CORESET that they are belong to?
    Can’t be search space with different sizes for same CORESET?”

The SearchSpace frequency and time resources are identified by the monitoringSlotPeriodicityAndOffset, duration and monitoringSymbolsWithinSlot. It doest not need to cover the whole CORESET area. Each SearchSpace is individually defined.

2 Likes

So what’s the significant of CORESET #0 ? is it also for giving info on pdsch and pusch data?
you said @marcengo the uE needs to decode and obtain the controlResourceSetZero and searchSpaceZero as first thing for decoding coresets …so why not starting from coreset #1 or coreset #6 ? just Im trying to understand the implicitly meaning and significant of CORESET0 from gNB side and from uE side.

You mean by saying - “information” - in "for the porpoise of decoding common and UE-specific information " … is the common/specific data of pdsch, pusch ?

So if I understand then in the searchspace there might be overlapping arreas of aggregation level 2 to 1 or 4 …etc? means searchspace can have multiple aggregation level scattered in the searchspace without any consideration to the overlapping places of different aggregation levels allocations within that searchspace? for example assume search space with two rbs so assume in that searchspace I have aggLvl1 and aggLvl2 so here as u see we have overlapping between the places of aggLvl1 and aggLvl2 .

Number of candidates are actually the specific places within searchspace correspond to his aggLvl where my uE will search for its dci?

So you mean like each searchspace size is configureable individually ? so it would be like according to searchspace index domain we can configure under that searchspace index its size and its parameters…right?

So the search space area/place of uE specific and common can be overlapped and if so … is it called one search space or two searchspaces ? let me example it
assume I have search space from rb index 0 - 100 , so assume uE specific search space is from 0-100 , and common search space is from 0 - 50 so lets say in the same subframe I’ve common and uE specific … as u see they are overlapping … so we call them as one search space or two search spaces ? (because already rbs 0-100 are occupied by uE specific search space and from 0-50 we also have common search space …but implicitly 0-50 also occupied for uE specific search space so … “overlapping” within same subframe Im talking … )

Here you mean according to release 15 at different bandwidth parts we must have different coresets ids? it means if I used in BWP #1 coreset indexes 0 1 2 so in BWP #2 coreset indexes must be totally different from BWP #1

Right?

In certain BWP ; Per slot -subframe- I can have one contagious search space Yes? mean the search space can’t be non contagious in frequency within certain BWP.

@marcengo I appreciate you much for taking time marcengo and I hope also day will come to help others here. MUCH THANKS !

1 Like
  • Imagine the NR channel as a town with hundreds of houses (resources).

  • 1 or few of these houses are designated as coresets.

  • The different rooms in any coreset house are the CCEs.

  • If you want to search for something in the whole house, it will take a long time.

  • To reduce search time, 1 or more search areas inside the house to be defined based on what you are searching for.

  • For example, if you are looking for a pen, you can search in rooms 1/2/3 but if you are looking for a bottle, you can search in rooms 1/4/7/10 and so on. Note that some rooms can be part of different search spaces

1 Like

I understand your analogy -really make more sense now, appreciated.
just to be clear , about saying “Note that some rooms can be part of different search spaces” you mean like in your example the search of pen and search of bottle are sharing the same room ’ 1 ’ … so room 1 is part of different search spaces ; Correct?

Yes that’s correct

1 Like

What about the aggregation level and candidates …according to your analogy … how I can imagine them? thanks alot for clarifications !

Well Aggregation levels concept may not perfectly fit in this analogy but let me try.

AL means that the thing you are searching for may not be in a single room but could be split on 2,4,8 or 16 rooms.

If you don’t know then you have to search for all room combinations. To speedup the process, you may be given a hint on which combinations of rooms which are the candidates

For example, if your search space consists of 8 CCEs
AL8 has 1 candidate (0,1,…,7)
AL4 has 2 candidates (0,1,2,3) & (4,5,6,7)
AL2 has 4 candidates (0,1)(2,3)(4,5)(6,7)
AL1 has 8 candidates 0,1,2,3,4,5,6,7

In this case we have 15 valid candidates but the network can tell the UE that this search space has AL4 and AL8 only (no AL1 and no AL2) so the UE will have to search in 3 candidates only (in the worst case) and hence speeding up blind decoding process

1 Like

I understand , thanks much and appreciated!

per slot -subframe- within active BWP , how many coresets I can have?
what Im trying to understand lets assume I have three different coresets that each one has 3 symbols in time .
So in the slot , the three different coresets will be distributed along 14 ofdm symbols that’s found on slot? or the meaning of three different coresets mean that over same “3 ofdm symbols in time” will have all the three coresets distributed in frequency division(means they have same time -same 3 ofdm symbols- but divided in frequency along active BWP) ???

thanks.

@Ryan I’ll try to share my understanding and hope it answers your question

At least in Release 15, there are a maximum of 3 coresets within BWP.

In the frequency domain, the coresets spans multiples of 6 RBs

In the time domain, the coreset does need to be available in every slot. It could be repeated every X slots

Within one slot, there might be multiple instances of the same coreset in the time domain. The main restriction is that the duration of each instance is fixed and up to 3 symbols. For example, the “same coreset” may exists in symbols 0,1,2 and 7,8,9 within the same slot.

1 Like

I understand. So as u said -
“there might be multiple instances of same coreset in the time domain” … so can’t be different coresets in the same slot?

another question, so in frequency domain can be per active PWB three coresets in one slot ? assume that this active PWB support good amount of rbs …

let me draw what’s Im trying to say - (in my photo coreset defined as “core”)

Here in this photo 1 active BWP and three coresets …is it valid/possible to have coresets like this?

Now the second confusion is this -(in my photo coreset defined as “core”)


Here in this photo above 1 active BWP and three coresets …is it valid/possible to have coresets like this?

third confusion is this -(in my photo coreset defined as “core”)


Here in this photo above 1 active BWP and three coresets …is it valid/possible to have coresets like this?

Appreciated to explain and clarify my confusions / questions here !

MUCH MUCH thanks and appreciated.

I believe all the three figures are generally valid.

However for the second case, since the 3 coresets have same frequency domain and same duration 3 symbols, they could belong to one coreset

I understand , Yes it’s now more make sense.

MUCH appreciated ! ! !

Just Im curious and wondering like lte we have dynamic cfi/dcfi
(increasing cfi symbols from 1 to 2 or to 3 symbols)…so we can have implicitly dynamic coresets in 5g nr? if so then how just conceptually we can implement that? just to grasp the overview idea how we can do same lte concept of cfi/dcfi to coresets.

The coreset size is fixed per my understanding right? so can’t be dynamically like cfi in lte?

Another question about searchspace that has different multiple aggregation level …it means for each uE will have to search in all aggregation level that’s defined in the searchspace AL 2 , AL4 , AL8 , AL16? so there will be overlapping between places of aggregation level within the searchspace as Im showing here down -

Am I right as what I shown on the photo? each searchspace has defined aggregation level with defined candidates(n2)…so as u see may there will be overlapping in the places of aggregation level within the searchspace …

Another question - within certain searchspace there can be found a multiple of different candidates with different aggregation level inside that searchspace?

A pleasure to clarify with a good explanation if there’s something wrong here in the photo

@O.B
@marcengo

Any help please? appreciated.

It was great discussion regarding PDCCH resource allocation, Thanks all for sharing the information and knowledge.