Machine Learning Overview and its Uses in Telecom

Today we are discussing about the very Hot topic for nowadays  which is Machine Learning. First we will find what is Machine Learning, why we need Machine Learning ,Types of Machine Learning and how Machine Learning can help Telecom.

What is Machine Learning?- Idea behind the machine learning is to make machines(computer) capable of taking decision without explicitly programmed means once machines are trained and new data comes in the system then machine can train themselves without reprogrammed again.

Why we need Machine Learning?- There are lots of data generates in every field nowadays and that data too is present in the form of structured and unstructured. With processing by Humans for these data are not so easy so these data can be used for training the machine and machine can take efficient and fast decision. Machine is also do fast computation and complex thing which human cant do.

How does Machine Learning Work?- Machine Learning is a technology in which machines learns through training( We have to train a model with training data sets often known as inputs) and processes specific input to predict an output. In order to get desired output we have to properly trained the model with help of some parameters/Algorithms.

Uses of Machine Learning in Telecom or how can Machine Learning will help Telecom?- The telecom operators have already the masses of data in terms of its customer data, Network performance data, network traffic data and social media data. So there are some Machine Learning applications which are helping telecom:-

1. Using Machine Learning to identify restart the sleeping cells:-  In this application we can analyse from network performance data to identify the sleeping cells and trigger an automatic restart.

2.To identify potential churners:- The customer churn is highest in the Telecom in comparison to other fields, so for the telecom operators it is very important to retain there customers. With help of Unsupervised Learning( We have customer behaviour data is already with us) we can identify the potential churners and plan accordingly.

3.Using Machine Learning to improve customer service application.

4.Using Machine Learning to identify fraud mitigation.

There are different types of Algorithms used in machine Learning as listed below. We will discuss each one in next blog.

What is NB-IoT(Narrowband IoT), its operation mode and features

In order to meet the requirements of IoT world, the 3GPP has designed the narrow band internet of things (NB-IoT) standard in its recent Release 13. The NB-IoT operates on the principal of LPWA(Low power consumption and wide coverage).

Uses of NB-IoT:- The potential uses of NB-IoT is communication(Machine 2 Machine) for IoT applications. Some of are as below:

1.Smart Metering(Electricity, Gas, Water)

2. Smart City

3. Security systems

4. Asset Tracking

5. Agriculture

Operating Mode for NB-IoT: NB-IoT is working on 180Khz Bandwidth(1 PRB in LTE).There are 3 operating modes which are explained below:

1. In-Band Operation:- Utilizing Resource blocks within an LTE carrier.( No Extra cost for Hardware upgradation but chances of increasing interference is high as we are using 1 PRB within the same band).

2. Guard-Band Operation:- Utilizing the unused resource blocks within an LTE carrier guard bands.(No Extra cost for hardware as well as chances of Interference is very less)

3. Standalone Operation:- Utilizing of currently used GSM frq.( Hardware upgradation required and chances of Interference is more, we have to do spectral refarming to overcome this).

Features of IoT:- There are 2 important features of NB-IoT which are listed below.

1. Wide Coverage Area:- NB-IoT offers 20dB of MCL(Maximum coupling loss) higher than LTE(up to 164dB). In order to meet this several modification have been deployed and one of the major is number of retransmissions(128 for Uplink and 2048 for Downlink). These repetitions are combined at receiver side to improve the SINR.

                                 There are 3 Coverage Enhancement(CE) mode in NB-IoT.

CE mode0:- Normal coverage with MCL~144dB and sub carrier spacing of 15khz.

CE mode1:- Robust coverage with MCL~154dB and  sub carrier spacing of 15Khz.

CE mode2:- Extreme coverage with MCL~164dB and sub carrier spacing of 3.75Khz.

The choice of the mode dependent upon channel conditions. Each mode determines the transmission parameters including the number of repetitions. Depending upon the coverage level the serving cell indicates the UE to repeat the transmission(in UL and UE informs to eNodeB in DL) using the same transmission power on each transmission. Combining the different retransmissions results in coverage enhancements.

2. Low Power consumption:- there are mainly 2 techniques used in NB-IoT to achieve this.

       Idle mode extended discontinuous reception (I-eDRX): this mode allows a discontinuous reception for maximum of 3 hours, which saves UE battery but still allows it to be reachable by the network through paging messages or downlink control channels.
Power saving mode (PSM): this power saving mode allows unconnected state for up to 13 days, where UE enters to a deep sleep. Unlike in I-eDRX, UE is unreachable while remains registered in the network.

The configurations and implementation of NB-IoT will discuss in next blogs..

Tracking area update(TAU) in LTE

Before jumping to Tracking area update procedure lets first understand what is Tracking Area(TA) and why we need TA(Tracking Area).

What is Tracking Area?- Tracking Area in LTE is geographical combination of several eNodeBs.
                       Each TA has two main identities:- Tracking Area code(TAC) and Tracking Area Identity(TAI).
    TAI= PLMN ID+TAC.

Why we need Tracking Area?- In Idle when there is no signalling and data bearers are associated with it the location of UE to MME is known by Tracking area(TA) only.

Now as we know what is TA and need of TA we are ready to discuss Tracking are update.

Tracking Area Update:- Tracking area update initiated by UE(direction is from UE to NW) only after the UE completes the attach procedure and moves into the EMM(EPS Mobility Management) state.

Note:- UE can trigger TAU in RRC idle state or in connected state but procedure ends only when the UE is in RRC connected state.

Reasons for TAU?- There are many reasons for Tracking Area Update(TAU). Some are listed below:-

> When a UE moves to a new Tracking Area.
> When the Periodic update timer T3412 expired. The value of T4312 is in the attach accept message.
> Registering with the Non EPS services(CS Domain) when the UE is already attached for EPS services. This includes completing an IMSI attach as part of TAU procedure.
> Registering for an EPS services after an inter RAT changed( From GERAN/UMTS to LTE).
> Re-Registering to LTE after CS fallback.
> In case of MME load balancing( UE initiates TAU procedure when eNodeB releases the RRC connection with cause "Load Balancing TAU required".

Note:- EPS update type should be as below:
"TA updating", "Combined LA/TA updating", "combined LA/TA updating with IMSI attach", "Periodic updating".

         

TTI Bundling in LTE

TTI Bundling:- Today we are going to discuss an LTE feature to improve mainly VoLTE performances at cell edges or in poor radio conditions.

Need to TTI Bundling?- TTI bundling as its name suggest is a bundling(together as a package) of TTI(smallest unit of time in which eNodeB is scheduling the UE).

       In normal scenario(without TTI bundling) at cell edges or in poor RF condition, the UE informs eNodeB about its power getting maximum value (UE has limited power 23dBm in LTE) therefore the chances of re-transmission of Voice Packets is increasing(Multiple NACK). Which further leads in delay of VoIP packets arriving(It is not acceptable in case of Voice call).

Note: Normally 1 HARQ retransmission is taking a delay of 8ms.

How it Works?- In TTI bundling we are sending a transport block multiple times in a consecutive subframe(Normally 4) without waiting for HARQ ACK/NACK messages. A combined ACK/NACK can be sent after processing all the transmissions of a transport block.

                     Actually, in TTI Bundling, we are sending multiple redundancy(4) versions of the same set of bits in consecutive(Bundled) TTI and eNodeB sends back ACK when it successfully decodes the data. This way we are avoiding delay (Which is mainly due to RTT of HARQ).

Note: TTI Bundling uses QPSK modulation.

SRVCC in LTE

SRVCC( Single Radio Voice call continuity):-  SRVCC is a feature/process by which a voice call(VoIP) in LTE can be continued in legacy RAT(2G/3G, CS call) via handover when UE moves out of LTE coverage. 

Why its called Single Radio?- The process is called single radio is due to only a single radio is required in the handset.

Call Flow for SRVCC:- Here below I m indicating the simplest way of SRVCC call flow.

The process:- When UE moving away from LTE coverage then UE notifies the eNodeB(By sending measurement reports). Then eNodeB informs MME to make a decision for handover to legacy RAT network.

We are defining thresholds of SRVCC for serving cell by Inter-RAT B1 and B2 thresholds.

Note: We can set a greater threshold in case of TTI bundling feature is enabled. As with TTI bundling feature VoLTE services can be improved at cell edges.

https://mytechnew2019.blogspot.com/2019/02/tti-bundling-in-lte.html

New Interface: For SRVCC implementation  3GPP introduces a new interface called Sv. The Sv interface is from MME to MSC.

Elements of SRVCC Network Architecture:- Here we are discussing about some important network elements which play a big role in SRVCC.

Sv interface:- Signalling interaction for SRVCC handover as an interface between MME and MSC server.

eNodeB:- SRVCC indication to MME with help of UE measurement reports.

MME:- Bearer splitting for Voice and Non-Voice services and MSC selection.

Types of DCI formats in LTE

DCI(Downlink control information) :- It tells the UE how to get its data which is transmitted on PDSCH in the same subframe. Means with the help of DCI only UE can decode the PDSCH from the resource grid and find its data.

     The DCI formats provides the UE details such as number of resource blocks, resource allocation type , Modulation scheme which UE has to use(QPSK,16QAM,64QAM), transport block size, redundancy version, coding rate( Amount of information bits over transmitted bits), For MIMO configuration PMI and RI.

There are 13 DCI formats but we are discussing only below important formats.

Format Name                                Use
DCI-0                                     UL Grant(Scheduling of PUSCH)  
DCI-1A                                  Used during the Random Access Procedure
DCI-1                                     Scheduling of the DL PDSCH for TX diversity
DCI-2A                                 Scheduling of DL PDSCH in Open loop MIMO(TM3)
DCI-2                                    Scheduling of DL PDSCH in closed loop MIMO(TM4)                                            

Capacity of Volte User per cell

Now days there is massive shift for voice call to VoLTE calls from conventional CS calls. Here i am discussing about Max theoretical VoLTE User per cells.

For this first we have to calculate VoLTE call Packet size and Radio resource required for VoLTE packets . 

VoLTE call packet size mainly depends on below factors:
     > Codec used by UE
     >  UE Radio conditions
     > EnodeB Scheduling Algorithm
     > Protocol used.
Lets try to find out the VolTE call packet size with below assumption.

Assumptions:-
LTE System Bandwidth: 5MHZ(25 PRB)
Duplex Mode: FDD
MIMO: 2*2
Codec used: AMR-Wide band codec(12.65)
RoHC: Enabled.

Calculation: AMR-WB(12.65) coder generates about 253 bits of speech every 20ms( Voice data is genearted in every 20ms for VoLTE). In order to deliver each voice samples to UE needs to add protocol headers such as RTP header(12 bytes), a UDP header(8 bytes) and an IP header(40 bytes).

        So VolTE packet length with all headers are:
 =Codec bits + RTP Header + UDP Header + IP Header
 = 253bits + 12*8(=96bits) + 8*8(=64 bits) +40*8(=320bits) = 733bits/20ms.

Note: When RoHC is enabled it will replaced the RTP, UDP and IP header with a much smaller RoHC header(24 bits).

Further RLC and MAC layers will add their own headers hence approx. 300 bits are needed for every VoLTE voice call.

Radio resource calculation for VoLTE packets: In LTE one PRB has 12 subcarrier and 14 symbols(Normal CP) over 1ms duration. 
 Total number of REs over 1ms = 12*14=168 but some of the REs are used by control symbols(PDCCH and RS). So final count is around 120 REs.

LTE DL modulation supports QPSK, 16QAM, and 64 QAM for PDSCH which means each resources can carry 2 bits, 4 bits and 6 bits. But some of these bits used for error control also.

Consideration: Lets consider CQI 15= Good ,provides 64 QAM, CQI 7= Average ,provides 16 QAM and CQI 1= Poor, provides QPSK.

When CQI =15:-
 Modulation: 64 QAM(6 bits)
 Effective coding rate = 0.926
 Total data bits in 1 RE = 6*0.926 = 5.55
 Total data bits in 1 PRB = 120*5.55 = 666 data bits or equivalent to 2 VoLTE calls.
But LTE schedular can not allocates < 1PRB so 1 PRB is needed for VolTE calls with CQI=15.

When CQI = 7:-
Modulation: 16 QAM(4 bits)
 Effective coding rate = 0.369
 Total data bits in 1 RE = 4*0.369 = 1.476
 Total data bits in 1 PRB = 120*1.476 = 177 data bits. So 300 VolTE bits required 2 PRB.

When CQI =1:-
Modulation: QPSK(2 bits)
 Effective coding rate = 0.076
 Total data bits in 1 RE = 2*0.076 = 0.152
 Total data bits in 1 PRB = 120*0.152 = 18 data bits. So 300 VolTE bits required 16 PRB.

Now we are ready to calculate max theoritacal VoLTE user per cell:- In VoLTE voice data is generated every 20ms so if everything is good then about 20 VoLTE calls can share the same set of PRB one after another.

The max VolTE call that can be carried = (No of avg PRB)/ No of PRB required per VoLTE calls *20.
Hence VolTE calls per CQI and Bandwidth is as below:

Bandwidth             1.4MHZ         3 MHZ         5MHZ             10MHZ             15MHZ           20MHZ
No of total PRB         6               15               25                     50                   75                100
CQI15(1 PRB)         120             300              500                 1000                 1500            2000
CQI7(2 PRB)            60              150              250                  500                  750             1000
CQI1(16 PRB)           8                18               31                    63                    94               125

The above is the theoretical max value but in practical there are some issues. We will discuss it in next..     

Advantage of 5G NR(New Radio) over LTE

Recently 3GPP has shared the Non-stand Alone 5G NR(New radio) in Rel15. I would write an article on Advantage of 5G over legacy LTE.

Major Goals for 5G:- There are many targets that need to achieve in 5G NR but the most important is higher throughput(in Gbps) and low latency(<1ms). 

          Below are the some features which makes 5G better than LTE.

eMBB(Enhanced Mobile Broadband):- Just like 4G where we have eNodeB in 5G NR we have eMBB. eMBB is the throughput oriented part of 5G NR. It has following characteristics:-

    1. Peak Data Rates- 10 to 20Gbps
    2. 10000 times more traffic.
    3. Supports macro and small cells.
    4. Supports high mobility of about 500Kmph.
    5. Helps in Network Energy saving by 100 times.

URLLC( Ultra reliable low latency communication):- In 5G NR E2E latency is <1ms while in LTE having around 20-100ms  and sometimes >100ms in case of congestion. To decrease/Improve the latency in 5G NR we have a dynamic structure of carrier spacing where symbol length can change along with the slot length.

   carrier spacing supported in 5G NR are: 15,30,60,120 and 240Khz.

In LTE there are fixed two slots/Subframe(0.5ms) but in 5G NR no. of slots may vary with carrier spacing( max upto 16 slots/subframe with 4.4microseconds).

Bandwith:- 5G NR supports a single carrier of 275RBs compared to 100 RBs in LTE.

Massive MIMO:- Greater than or equal to 8*8 MIMO we can implement in 5G NR.

Asynchronous HARQ:- Since we are using a flexible symbol length in 5G NR it also introduces a HARQ PDSCH feedback timing indicator. Consider a below example:
   Since we have a slot length of 0.0625ms and suppose UE sends the ACK in N+4 slot then the overall time would be 0.0625*4=0.25ms. 
So above example will ensure that the URLCC target is achieved.

ENHANCED PDCCH LINK ADAPTATION IN LTE

Here we will discuss about a LTE Feature called Enhanced PDCCH Link Adaptation. The main aim of this feature is to increase the capacity to allow more users to be scheduled by improving the efficiency of PDCCH usages. The capacity can be improved by Utilization of CCE in more efficient way.

How it Works: In every period channel quality(CQI) is reported to eNodeB by UE. In basic PDCCH Link adaptation its uses the PDSCH successful transmission and its offset as an information for PDCCH to achieve any particular BLER.
                          Enhanced PDCCH Link Adaptation uses PDSCH HARQ-ACK from UE and also the PUSCH detection result to determine whether the PDCCH transmission is successful or not.

Algorithm Flow:

 1. SINR Estimation: Based on the CQI Report report that UE sent to eNodeB.

 2. PDCCH Outer Loop Adjustment: To Achieved the required BLER.

 Elements to determine the criteria:
   > Success or failure of the PDCCH transmission. The HARQ-ACK reports and PUSCH Detection results are used for this. As long as the eNodeB can detect the ACK, either positive or negative, PDCCH transmission considered successful. 

> Aggressive or Conservative PDCCH Allocation. The PDCCH allocation is considered conservative if the allocation of CCE exceeds the required number and the aggressive one is the other way around.

From the outer loop adjustment, the output is required number of CCE for the same.

 The result of the implementation can be seen in the below chart.

Based on the chart above it can be seen that the CCE used decreased by 27% and the cell throughput improved by 1.55%.

Conclusion: This feature is capable to occupy CCE more efficient which lead into the increased cell Throughput.