Networked-Signal Processing in OAI

• Address the trend to split/slice/distribute the radio and protocolstack between network entities1. spatial signal-processing (distributed MIMO)2. network function virtualization3. data-center processing

• current implementation and what’s coming• some notes on EURECOM deployment• some generic information from OAI community

Raymond KnoppOpen5G Lab, EURECOMSophia Antipolis, FRANCE

04.02.2017, FOSDEM(Brussels)

NGFI Harmonization in OAI

• New descriptions for OAI RAN infrastructure Node Functions

– NGFI RCC : Radio Cloud Center– NGFI RAU : Radio Aggregation Unit– NGFI RRU : Remote Radio Unit– 3GPP BBU : Baseband Unit– 3GPP eNodeB : Complete eNodeB

Source: China Mobile - http://www.windriver.com.cn/windforum/download/wf2015 networking03.pdf

• Current OAI implementation(RRU/RCC) supports either– IF5 time-domain

fronthaul (¿ 1 GbErequired)

– IF4.5 split (FFTs) (280Mbit/s/antenna portfronthaul 20 MHzcarrier) per carrier/sector

– Soon IF2 (NFAPI) IF1for super-PDCP soon

NGFI split points

Functional Splits

MACTX

PDCPTX

RLCTX

MACRX

PDCPRX

RLCRX

RRC

GTP-C GTP-U

PHYTX

PHYRX

rf device

MACTX

PDCPTX

RLCTX

MACRX

PDCPRX

RLCRX

RRC

GTP-C GTP-U

PHYTX

PHYRX

if device

3GPP eNodeB 3GPP BBUto MME to S-PGw to MME to S-PGw

USB3 or

PCIeNFGI IF5 /

ethernet

MACTX

PDCPTX

RLCTX

MACRX

PDCPRX

RLCRX

RRC

GTP-C GTP-U

upper-PHYRX

if device

NGFI RCC (IF4.5)to MME to S-PGw

NFGI IF4p5

/ ethernet

upper-PHYTX

to NGFI RRUto NGFI RRUor RRH gw

to RF Device

MACTX

PDCPTX

RLCTX

MACRX

PDCPRX

RLCRX

RRC

GTP-C GTP-U

upper-PHYRX

if device

NGFI RCC (IF4p5/IF1”)to MME to S-PGw

NFGI IF4 IF1pp

/ ethernet

MACTX

PDCPTX

RLCTX

MACRX

PDCPRX

RLCRX

RRC

GTP-C GTP-U

if device

NGFI RCC (IF1”)to MME to S-PGw

NFGI IF1pp /ethernet

PDCPTX

PDCPRX

RRC

GTP-C GTP-U

if device

NGFI RCC (IF1’)to MME to S-PGw

NFGI IF1p /ethernet

to NGFI RRU

to NGFI RAU

to NGFI RAU

Functional Splits

upper-PHYRX

if device

NGFI RAU(IF1”)

NFGI IF4p5

/ ethernet

upper-PHYTX

if device

NFGI IF1pp

/ ethernet

to NGFI RCC

to NGFI RRU

upper-PHYRX

if device

NGFI RAU(IF1’/IF4p5)

NFGI IF4p5

/ ethernet

upper-PHYTX

if device

NFGI IF1p /

ethernet

to NGFI RCC

to NGFI RRU

MACTX

RLCTX

MACRX

RLCRX

upper-PHYRX

if device

NFGI IF4p5

/ ethernet

if device

NFGI IF1p /

ethernet

to NGFI RCC

to NGFI RRU

MACTX

RLCTX

MACRX

RLCRX

NGFI RAU(IF1’/IF4p5/IF1”)

Functional Splits

if device

NGFI RRU (IF5)

NFGI IF5/

ethernet

rf device

NGFI RRU (IF4.5)

to NGFI RCC,NGFI RAU,3GPP BBU

to RF Device

USB3,

PCIe,

CPRI

if device

NFGI IF4p5/

ethernet

rf device

to NGFI RCC,NGFI RAU

to RF Device

USB3,

PCIe,

CPRI

lower-PHYRX

lower-PHYTX

if device

upper-PHYTX

if device

NFGI IF1pp 4P5

/ ethernet

to NGFI RCC, NGFI RAU

USB3,

PCIe,

CPRI

to RF Device

Functional Splits

NGFI RRU(IF1”/IF4p5)

lower-PHYRX

lower-PHYTX

Some Notes on usage of splits

OAI RAN Software Architecture

PHY Instance 0

PHY Instance 1

MAC/RLC/PDCP Instance 0DL/UL

Scheduler

MACRX

PDCPRX

MACTX

PDCPTX

1ms TICK

DCI0 , Transport Blocks

DCI1 DCI 2

RLCRX

RLCTX

PHY RX neNB0

PHY RX neNB1

PHY RX neNB2

PHY TX n + 4eNB0

PHY TX n + 4eNB1

PHY TX n + 4eNB2

Transport Blocks

Transport Blocks

CQI/SR/ACK/NAK/PHRTransport Blocks

CQI/SR/ACK/NAK/PHRTransport Blocks

CQI/SR/ACK/NAK/PHRTransport Blocks

1ms TICK

DCI0 , Transport Blocks

MAC/RLC/PDCP Instance 1

MACTX

PDCPTX

RLCTX

MACTX

PDCPTX

RLCTX

MACRX

PDCPRX

RLCRX

MACRX

PDCPRX

RLCRX

PHY RXeNB0

PHY TXeNB0

CQI/SR/ACK/NAK/PHRTransport Blocks

DL/ULScheduler

MACRX

PDCPRX

MACTX

PDCPTX

RLCRX

RLCTX

RU/eNodeB Instances and Component Carriers

RU3

RU0

RU1

RU2

RU4

RU5

RU6

• Radio Unit (RU) is

– an entity managing a set of physical antennas. It can have a local RFunit or remote RF unit

– performs precoding of multiple eNB TX streams and OFDM modulation(TX) and demodulation (RX) (part of 36.211)

• eNB Instance (indexed by Mod id, or enb mod id) is a separate set ofthreads and contexts for a full eNodeB, or at least the MAC/RLC entities,in the same Linux process. There is one MAC/RLC entity associated to allcomponent carriers of a single eNB instance.

• eNB Component Carrier (indexed by CC id) is

– a software entity managing the L1 procedures (36.213,36.212,36.211)and can act on∗ sectored antenna component∗ Rel10+ component carrier∗ virtual cell for DAS or Massive-MIMO array

– each eNB is managed by one or two threads which operate on asubframe (TX and RX) and can have a local RU or remote RU

– if a remote radio unit the eNB performs the 36.213 specifications only(HARQ, etc.) and connects to the remainder via the IF2 xhaul interface.

RU/eNodeB Instances and Component Carriers

RU/eNodeB Instances and Component Carriers

• RU may have both an if device for fronthaul and an rf device forinterconnection with a local RF unit

• if the rf device is absent, it must have a southbound fronthaul interface(either IF5 or IF4p5) depending on the local processing of the remote RU

• if the if device is absent, it must have a southbound RF interface andrf device.

• three types of L1 processing are performed by the RU

– subset of common L1 procedures from 36.211 specifications– fronthaul compression/decompression– framing

• on TX

– A-law compression for (NGFI RAU IF4p5, NGFI RAU IF5)– A-law decompression (for NGFI RRU IF4p5 and NGFI RRU IF5)– OFDM modulation and cyclic prefix insertion (for

NGFI RRU IF4p5,NGFI RAU IF5,3GPP eNodeB BBU,3GPP eNodeB)– Precoding (for NGFI RAU IF5,

NGFI RAU IF4p5,3GPP eNodeB BBU,3GPP eNodeB)

RU/eNodeB Instances and Component Carriers

• on RX

– A-law compression for (NGFI RRU IF4p5, NGFI RRU IF5)– A-law decompression (for NGFI RAU IF4p5 and 3GPP eNodeB BBU)– cyclic prefix removal, frequency-shifting, OFDM demodulation, PRACH

DFT (for NGFI RRU IF4p5, NGFI RAU IF5, 3GPP eNodeB BBU,3GPP eNodeB)

• IF5 transports packets of size equal to a subframe and corresponding to a 1ms chunk of signal in thetime-domain. This is done via the functions send if5 and recv if5, in the layer1 transport procedures(openair1/PHY/LTE TRANSPORT/if5 tools.c). A timestamp is given along with the samples, corresponding tothe time (in samples) of the first sample of the packet.

• each block can be compressed with A-law compression, yielding a compression rate of .5.

RU Fronthaul interfaces (NGFI IF5)

PHY RX neNB2 (211,212)

PHY TX n + 4eNB2 (211,212)

RU0

PHY RX neNB1 (211,212)

PHY RX neNB0 (211,212)

PHY TX n + 4eNB1 (211,212)

PHY TX n + 4eNB0 (211,212)

RU->ru time.rxdata[0]

RU->ru time.rxdata[R-1]slot fep ul

36.211

eNB->common vars.rxdataF[0]

eNB->common vars.rxdataF[R-1]

Ala

w−

1()

from IF5 if device

PRECODING

RU->ru time.txdata[0]

doofdmmodrt()

36.211

openair1/SCHED/ru-procedures.c

Ala

w()

to IF5 if device

RU->ru time.txdata[T-1]

recv IF5()

send IF5()

ru fep full()

IF5 RU eNB end

7.5

kHz()

ru thread()

ru thread asynch rxtx()

• IF4p5 transports packets of size equal to an OFDM symbol (for DLRE and ULRE) indexed by the symbol,subframe and frame number. This is done via the functions send if4p5 and recv if4p5, in the layer1 transportprocedures (openair1/PHY/LTE TRANSPORT/if4 tools.c).

• each block are compressed with A-law compression, yielding a compression rate of .5.

RU Fronthaul interfaces (NGFI IF4p5)

PHY RX neNB2 (211,212)

PHY TX n + 4eNB2 (211,212)

RU0

PHY RX neNB1 (211,212)

PHY RX neNB0 (211,212)

PHY TX n + 4eNB1 (211,212)

PHY TX n + 4eNB0 (211,212)

Ala

w−

1()

from IF4p5 if device

PRECODING

openair1/SCHED/ru-procedures.c

Ala

w()

to IF4p5 if device

recv IF4p5()

send IF4p5()

IF4p5 RU eNB end

eNB->common vars.rxdataF[0]

eNB->common vars.rxdataF[R-1]

eNB->prach vars.prachF[0· · · R-1]

ru thread()

RU Fronthaul interfaces (NGFI IF4p5)

RU0

RU->ru time.rxdata[0]

RU->ru time.rxdata[R-1]slot fep ul

36.211

from rf device

RU->ru time.txdata[0]

doofdmmodrt()

36.211

openair1/SCHED/ru-procedures.c

to rf device

RU->ru time.txdata[T-1]

ru fep full()

IF4p5 RU remote-end

Ala

w−

1()

Ala

w()

7.5

kHz()

recv IF4p5()

send IF4p5()

to if device

from if device

ru prach procedures36.211

ru prach procedures()

rx prachl36.211

ru thread prach()

ru thread()

ru thread asynch rxtx()

to if device

TX Precoding

• Spatio-temporal filtering for muli-cell (vCell) and multi-usertransmission. Input and output are frequency-domain signals.

• can be applied to Rel-10/11/12/13 physical channels and Rel-8common channels

– UE-specific precoding (TM7-10)– vCell-specific precoding (PDCCH + TM1-6) for groups of UEs– PMCH vCells

• Precoding applicable to

1. indoor DAS2. outdoor co-localized arrays (e,g, Massive-MIMO)3. outdoor CoMP

UE

vCell

vCell

TX Precoding

TX Precoding (to RF device)

eNB[0]

eNB[1]

eNB[CC max]

rf device[0]

rf device[1]

rf device[N-1]

common signal precoding(vCell)

PBCH,PSS/SSS,PCFICH/PHICH/PDCCH

PDSCH - TM1-6

UE-specific signal

precoding(vCell)

UE-RS,PDSCH - TM7-10

PMCH precoding

ph

yva

rseN

B.t

xda

taF

ph

yva

rseN

B.t

xda

ta

doofdmmodrt()

36.211

TX Precoding (to IF device, NGFI IFv4p5)

eNB[0]

eNB[1]

eNB[CC max]

if device[0]

if device[1]

if device[N-1]

common signal precoding(vCell)

PBCH,PSS/SSS,PCFICH/PHICH/PDCCH

PDSCH - TM1-6

UE-specific signal

precoding(vCell)

UE-RS,PDSCH - TM7-10

PMCH precoding

NGFI IFv4p5

NGFI IFv4p5

NGFI IFv4p5

• On TX path– eNB instances/component carriers operate on a set of logical

antenna ports (0-3 for TM1-6, 4 for eMBMS, 5 for TM7, 6 forpositioning, 7-8 for TM8, etc.)

– each eNB has a list of RUs and the logical antenna ports aremapped to the physical antennas attached to the RUs via theprecoding function

– if the eNB has a remote RU then it is connected by an xhaulinterface (IF2) to the eNB instance/component carrier in anotherunit which contains the RU. Here the eNB processing is split.

• on RX the eNB instance/component carrier has access to all RUsignals for processing

RU/eNodeB Instances and Component Carriers

PHY RX neNB0 (211,212)

PHY TX n + 4eNB0 (211,212)RU0

RUN−1PHY RX neNB0 (213)

PHY TX n + 4eNB0 (213)

IF2 (TX)

IF2 (RX)

IF1” (TX)

IF1” (RX)

logical antennas (TX)

N physical antennas (RX)

openair2 MAC(36.321)

RCCelement

RAUelement

if device

NFGI IF4p5 / ethernet

upper-PHY TX(212,213)

if device

NFGI IF1pp

/ ethernet

if device

TX PRECODINGRX COMBINING

if device if device if device

• Example: RAU with NGFI IF1pp xhaul (MAC/PHY split) northbound,NGFI IF4p5 fronthaul southbound, 2 vCell logical interfaces (2 L1/L2instances, or 1 L2 instance and 2 CCs), 4 RRUs with NGFI IF4p5

upper-PHY RX(212,213)

upper-PHY RX(212,213)

upper-PHY TX(212,213)

RU0 RU1 RU2 RU3

eNB0 eNB1

RAU Example (DAS)

PHYRX0

if device

NGFI IF4p5 / ethernet

PHYTX0

if device

NGFI IF1p / ethernet

TX PRECODINGRX COMBINING

• Example: massive-MIMO RAU with NGFI IF1p fronthaul northbound, 8 L1component carriers,1 L2 instances, many local RRUs with NGFI IF4p5

southbound

if device if device if device

PHYRX1

PHYTX1

PHYRX7

PHYTX7

RLC

MAC

eNB0

eNB1

eNB7

RU0···15 RU16···31 RU32···47 RU48···63

RAU Example (Massive-MIMO)

OAI IF2 Interface

• OAI IF2 is the interface between the 36.213 Physical Layer Procedures (HARQ, SR, CSI, etc.) and thetransport/physical channel processing

• it can be networked, although this is not used as an xhaul interface at the moment. Considerations for usingFAPI/NFAPI P7 are underway.

phy procedures lte eNB TX phy procedures lte eNB RX36.213

PHY transport and physical channel procedures36.211/212

commonsignalprocedures()

pmchprocedures()

generatedlschparametersfromdci()

generateulschparametersfromdci()

phyconfigdedicatedstep2()

generatedcitop()

generatephichtop()

pdschprocedures()

pucchprocedures()

processMsg3()

rxulsch()

ulschdecoding()

extractcqi()

getMsg3allocret()

lteesttimingadvancepusch()

lteeNBI0measurements()

processHARQinfo()

generate dci top()

MAC

DCI PDU

dlsch coding()36.212

DLSCH PDU 0

DLSCH PDU 1 0

DLSCH PDU 1 1

generate pcfich()36.212,36.211

generate dci0()36.212

pdcch scrambling()pdcch modulationpdcch interleaving

36.211

dlsch coding()36.212

dlsch coding()36.212

dlsch encoding()36.212

dlsch scrambling()dlsch modulation()

36.211

dlsch encoding()36.212 dlsch scrambling()

dlsch modulation()[2-4 layers]

36.211dlsch encoding()

36.212

openair1/SCHED/phy procedures lte eNb.c:pdsch procedures()

openair1/SCHED/phy procedures lte eNb.c:pdsch procedures()

openair1/PHY/LTE TRANSPORT/dci.c

dlsch coding()36.212

MCH PDUdlsch encoding()

36.212

dlsch scrambling()mch modulation()

36.211

openair1/SCHED/phy procedures lte eNb.c:pmch procedures()

ULCNTL

36.213generate phich top() generate phich()

36.212,36.211

openair1/PHY/LTE TRANSPORT/phich.c

PHICH REs

PCFICH REs

PDCCH REs

PDSCH REs

PMCH REs

generate pilots slot()generate pss/sss/pbch()

36.211openair1/SCHED/phy procedures lte eNb.c:common signal procedures()

CS-RS/PSS/SSS/PBCH REs

openair1/SCHED/phy procedures lte eNb.c:phy procedures eNB TX()

eNB TX Procedures

eNodeB 3GPP orNGFI RCC IF4p5

IF1’’ split points

openair1/SCHED/phy procedures lte eNb.c:phy procedures eNB RX()

MAC

RACHPreamble

ULSCH SDU(UE id)

eNB common vars.rxdataF[0]

eNB common vars.rxdataF[R-1]

rx ulsch(UE id)36.211

LTE TRANSPORT/prach.c

openair1/PHY/LTE TRANSPORT/ulsch demodulation.c

lte eNB I0 measurements()

ulschextractrbssingle()

36.211

lteulchannelestimation()

36.211

ulschdetectionmrc()

36.211

frequencyequalization()

36.211

lteidft()

36.211

ulschXXXllr()

36.211

ulschdecoding()

36.212

ulschchannelcompensation()

36.211

ulsch decoding.c

ACK/NAK,RI,CQI

pucch procedures(UE id)36.211

rx pucchl36.211

ACK/NAK

SR

TRANSPORT/pucch.c

prach procedures36.211

openair1/SCHED/phy procedures lte eNb.c:prach procedures()

rx prachl36.211

E PRACH>Lmin36.211

LTE ESTIMATION/lte eNB I0 measurements.c

eNB PHY RX Procedures

eNB->prach vars.prachF[0· · · R-1]

rx ulsch(UE id)36.211

Rr,l = DFTNfft(rr,l � F7.5), r = 0, 1, · · · , R − 1, l = 0, 1, · · · , Nsymb − 1 (eNB common vars→rxdataF[][])

slot fep ul36.211

rr,lRr,l

Rext,r,l

Rext,r,l(n) = Rr,l(12firstPRB + n), n = 0, 1, · · · , 12NPRB − 1 (eNB pusch vars→ulsch rxdataF ext[][])

Hr,l

Hr,l = Rext,r,l �DRS∗l (cyclicShift, nDMRS(2) , nPRS), (eNB pusch vars→drs ch estimates[])

Rcomp,r,l

Rcomp,0,l

Rcomp,0,l = 1R

∑R−1r=0 Rcomp,r,l

Rcomp,0,l

rcomp,0,l

Rcomp,0,l(n) = Rcomp,0,l(n)Q8

(1

|H(n)|2+I0

), H(n) =

∑R−1r=0 Hr(n)

Rcomp,r,l = Hr �Rext,r,l2− log2 |Hmax|, Hr = 1

2(Hr,3 + Hr,10)

(eNB pusch vars→ulsch rxdataF comp)

λl

QPSK : λl(2n) = Re(rcomp,0,l(n)), λl(2n + 1) = Im(rcomp,0,l(n)) (eNB pusch vars→ulsch llr)

rcomp,0,l = IDFT12NPRB(Rcomp,0,l)

16QAM : λl(4n) = Re(rcomp,0,l(n)), λl(4n + 2) = Im(rcomp,0,l(n))

λl(4n + 1) = |Re(rcomp,0,l(n))| − 2|h(n)|, λl(4n + 3) = |Im(rcomp,0,l(n))| − 2|h(n)|

ulschextractrbssingle()

36.211

lteulchannelestimation()

36.211

ulschchannelcompensation()

36.211

ulschdetectionmrc()

36.211

frequencyequalization()

36.211

lteidft()

36.211

ulschXXXllr()

36.211

eNB ULSCH Demodulation

eNB PRACH Detection

Rr = DFTNPRACH(rr), r = 0, 1, · · · , R − 1 (lte eNB prach vars→rxsigF[])

Rcomp,r = Rr �X∗u[i], r = 0, 1, · · · , R − 1 (lte eNB prach vars→prachF[])

r839,r = IDFT1024(Rcomp,r

), r = 0, 1, · · · , R − 1 (lte eNB prach vars→prach ifft[])

rr

DFT

Rr

�

X∗u(0)

X∗u(1)

R ×IDFT1024

R ×IDFT1024

∑r | · |

2

∑r | · |

2

max peak in sizeNCS2 window,keep delay

max peak in sizeNCS2 window,keep delay

preamble energy list[]

preamble delay list[]

r839,r

• PRACH detection is a quasi-optimal non-coherent receiver for vectorobservations (multiple antennas)

• correlation is done in the frequency-domain, number of correlations (inthe example above 2) depends on zeroCorrelationConfig configurationparameter

• peak-detection (for delay estimation) is performed in each NCStime-window

�

∑r,l,i | · |

2<>σ2[dB]

eNB PUCCH Detection

rx pucch()36.211

slot fep ul36.211

rr,lRr,l

Z∗l e−j2πf/16,−3 ≤ f ≤ 2

�

�

Z∗l

H

� sgn(Re(·))sgn(Im(·))

PUCCH1 (Scheduling Request)

PUCCH1a/1b (ACK/NAK)

• PUCCH1 detection is a quasi-optimal non-coherent receiver (energydetector) for vector observations (multiple antennas) for schedulingrequest. Care is taken to handle residual frequency-offset.

• PUCCH1A/1B detection is quasi-coherent based on a rough channelestimate obtained on the 3 symbols without data modulation.

• In both cases, correlation is done in the frequency-domain

OAI IF1” Interface

• OAI IF1” is the interface between the 36.321 Medium-Access (MAC) Layer Procedures and the 36.213 PhysicalLayer Procedures

• it cannot be networked today, Considerations for allowing an xhaul networking of this interface are underway.

Random-Access Events36.321

PHY procedures36.213

initiateraproc()

terminateraproc()

cancelraproc()

fillrar()

phyconfigdedicated()

phyconfigsib1()

phyconfigsib2()

phyconfigsib13()

eNBdlschulschscheduler()

getdciisdu()

getdlschsdu()

rxsdu()

getmchsdu()

getueactiveharqpid()

lteesttimingadvancepusch()

lteeNBI0measurements()

SRindication()

PHY Config Scheduling Meas

• Threads (all in targets/RT/USER/lte-ru.c)

– ru thread: Thread per RU which sequentially performs∗ read from south interface (RF or IF fronthaul)∗ RX processing for subframe n (if necessary).∗ wakeup eNBs that are waiting for signal (if necessary)∗ wait for eNB task completion (if necessary)∗ do TX processing for subframe n+ 4 (if necessary). Note that this

can spawn multiple worker threads for very high order spatialprocessing (e.g. massive-MIMO or DAS for UDN)

∗ do outgoing fronthaul (RF or IF fronthaul)– ru thread prach: Thread for PRACH processing in remote RU (DFT

on RX, IF4p5 RRU)– ru thread asynch: Thread for asynchronous reception from fronthaul

interface (TX direction in RRU). Allows for some jitter so as not toblock the RRU real-time processing.

• Synchronization on fronthaul interface

– synch to ext device : synchronizes to incoming samples from RF orFronthaul interface using blocking read

– synch to other : synchronizes via POSIX mechanism to other source(other CC, timer) which maintains real-time.

RU Threads

• Threads (all in targets/RT/USER/lte-enb.c)

– multi RX/TX thread mode (optional)∗ eNB thread rxtx: 2 threads per CC/Instance which do both RX

procedures for subframe n and TX procedures for subframe n+ 4.One operates on even subframes, one on odd. This allows 1mssubframe processing to use multiple-cores.

– single RX/TX thread mode (default)∗ eNB thread single: Thread per CC/Instance which sequentially

· blocks on signal from RU· RX/TX processing for subframe n and n+ 4· signals completion to RU

– eNB prach: Thread per CC id/Instance for PRACH processing

eNB Threads

eNB Timing (multi-thread mode)

SF n + 1 SF n + 2 SF n + 3

n

n + 4

SF n + 4

• The current processing requires approximately 1ms peak in each direction (basically 1 core RX, 1core TX).The current architecture will work on a single core if the sum of RX and TX procedures is limited to 1ms.It can fit on 2 cores if the sum of RX,TX and PRACH is less than 2ms.

• three threads, RX-TX even, RX-TX odd and PRACH. RX-TX blocks until woken by the RU thread with anew RX subframe n that is linked to this eNB process. The RX-TX thread performs ue-specific processingfor subframe n and then TX common and ue-specific processing for subframe n + 4 (frequency-domaingeneration only). This insures the data dependency between TX n + 4 and RX n is respected. Theduration of this thread should be less than 2ms which can compensate some jitter on the RX processing.

SF n

LTE HARQ periodicity (FDD, TDD can be longer)

PRACH n n + 1

n

n + 5n+1

n − 1

n + 3n−1

n + 2 n + 6n + 2

RU thread

RX-TX thread (even)

RX-TX thread (odd)

PRACH

eNB Timing

RX processing subframe n (eNB thread rx)

• trx read(1 ms)

• slot fep for slots 2n and 2n + 1[phy procedures eNB common rx()]

• launch PRACH RX thread (subframe n)– prach procedures

• phy procedures eNB rx() (subframe n)

TX processing subframe n + 4(eNB thread tx)

• phy procedures eNB tx

• ofdm mod

• trx write(1 ms)