Hello 5G experts, particularly those familiar with Ericsson RAN.
What specific inputs does the MAC scheduler consider when allocating downlink resources? Please avoid generic responses like channel conditions, CQI, or SINR.
I’ve noticed that even when a UE reports a good CQI, such as 12 or 13, the UE sometimes receives fewer PRBs with a higher MCS or more PRBs with a lower MCS.
Can you explain how Downlink scheduling works on the gNB side, specifically in the Ericsson case?
Scheduler for one UE is based on available resources, retransmissions, how long data is waiting in the buffer, buffer size, priorities etc.
MCS is based purely on PDSCH BLER. SO scheduler will lower the MCS till it reaches a PDSCH BLER of 10%.
Also CQI is taken into account.
Can we get some resources / reference which describes these points with an example?
In my case I didn’t observed any retransmission during near that slot. We were doing continuously FTP downloading.
But still I think gNB will understand the conditions of downlink with CSI report and SRS in TDD case where it doesn’t share PDSCH BLER.
Is there any other way to understand the PDSCH BLER by base station?
I think gNodeB gets NACK and ACK so it can calculate BLER.
Imagine a faulty UE that reports weird CSIs, or the CSI reporting standard deviation of UEs.
Ok, so means scheduler also taking ACK / NACK count while giving resource.
I wanted to see strategy used by Ericsson base station for giving dl resources.
Right in that case it can mislead base station.
I admit I do not know particularly for Ericsson. But for Huawei it works like I described above.
One more thing: any UE that is (detected as) out of sync is not scheduled at all.
For example CQI 0 means often no scheduling at all.
It would be a waste of resources.
When you say out of sync from base station in that case is there any mechanism or phy layer which monitor by base station to declare it out of sync?
There is always UL SINR, there is CQI reported there is ACK and NACK ratio.
So gNodeB has means to conclude if a Radiolink is good or useless.
As per my understanding related to Downlink link adaptation (but this is not for Ericsson!):
This will maintain the PDSCH BRLER.
If MCS is high then the same TB size can be accommodated in less PRB as the code rate will be higher.
So gNB will transmit this data and expect ack/nack even if after Max retx data not decoded by UE so it means the transmitted data more than what UE is expecting then MAC scheduler need to downgrade MCS to accommodate TB size so this process will continue and maintain the BLER.
Thanks I agree, but the point is why UE doesn’t get High number of PRB with High MCS to get large size of TB if UE is reporting good CQI, PDSCH BLER is very less, no retransmission req / NACK sent by UE and also there are continuously Downlink data in buffer.
I am thinking, there are multiple UEs in live network so it’s not possible to give full RB for a single UE in a particular slot, but wanted to know how gNB Downlink scheduler calculate that x-number of PRB should be assign to this UE in particular slot with mcs-y.
I wanted to understand the algorithm / strategy it uses while allocating PRB, MCS, Number of layer etc. because finally these attributes will be accommodated in to transport block.
I am seeking for some reference if possible for Ericsson or atleast Nokia.
It is purely based on amount of data for each UE in the buffer, QCI for each UE and how long data is staying in the buffer.
If there are 100 UEs on the cell you can imagine none will get great throughput.
There are also parameters for scheduling like type: proportional fair, enhanced proportional fair, round robin, max C/I.
Scheduler algorithm is toughest algorithm.
Every OEM has different algorithms and hence you will find different throughput even in the same radio environment.
Yes… that’s why i specifically asked for Ericsson. 
Each vendor has specific scheduling algorithms and it is proprietary.
Betting is done on robust scheduling algorithm.
In case of multi UE, based on UE reporting, scheduling will be done.
It is not some thing with one parameter and scenario, scheduling is decided.
They are multiple factors. Same scheduling algorithm works differently with different multi UE at different location.
Which scheduler algorithm is being referred to here, MAC?
Yes, MAC.
