LTE femtocell density modelling. Michael Fitch Chief of wireless research Technology Services and Operations BT Adastral Park, IP5 3RE October 2014

Transcription

1 LTE femtocell density modelling Michael Fitch Chief of wireless research Technology Services and Operations BT Adastral Park, IP5 3RE October 2014

2 What is a femtocell? Internet LTE EPC Long Term Evolution Evolved Packet Core Broadband Remote Access Server Digital Subscriber Line Access Multiplexer BRAS DSLAM Home hub LTE femto cell BT broadband A femocell is a very small cell designed to cover a large house with broadband - and to about 30m outside the house for voice and web browsing. Typical transmit power is < 100mW at 2.6GHz.

3 Benefits of doing the modelling For BT, it gives guidelines on the conditions and confidence under which we can deploy femtocells, For the consumer, it gives confidence of the experience of voice and data Gives installation guidelines and shows the importance of having interference mitigation mechanisms in place

4 The applications we used in the modelling UE type Range from femtocell Location Max bit-rate needed Set-top Box (STB) Up to 8m Inside only 20 Mbit/s Voice (VoLTE) Up to 30m Inside and outside LoS 25kbit/s Tablet Up to 8m Inside only 4Mbit/s Smart phone (Googler) Smart phone Openzone / Hotspot Up to 8m Inside only 500kbit/s Up to 30m Outside only LoS 500kbit/s

5 The question What is the maximum density of femtocells that: will support the applications and give adequate inside-to-outside coverage and give an improved user experience

6 London Bayswater is a very dense area (10,000 premises in 1 km 2 ), and was chosen as the area to use in the model Number of BT served premises is 3062

7 So let us set up a model MATLAB was chosen for this task, because of the easy ability to run many times and collect statistics the structure of the coding using functions enables straightforward debugging and calibration the reading and writing of large files is uncomplicated The model is based on 3-D geometry and pathlosses

8 Arrangement of different UE types 3 1. Set top box (inside) 2. Voice (inside and outside) 3. Tablet with HD (inside) 4. Googler (inside) 5. Hotspot (outside only) Up to 8m (inside) Up to 30m (LoS)

9 Wanted paths are either totally inside or inside-tooutside 3 1. Set top box (inside) 2. Voice (inside and outside) 3. Tablet with HD (inside) 4. Googler (inside) 5. Hotspot (outside only) Wanted paths Up to 8m (inside) Up to 30m (LoS)

10 Interference from neighbouring femtos is inside-to-inside or inside-to-outside Unwanted inside to inside interference (nlos and LoS) Unwanted in to outside interf (nlos and L 1 4 Wanted paths Up to 8m (inside) Up to 30m (LoS)

11 Outline method of modelling outage probability Femtocell locations are chosen at random from the test area, with a density of 0.01 to 1, The horizontal position is dithered, otherwise all femtocells would be in the exact centre of the property, One UE of each type is placed around each femtocell, 1.5m from the floor For each UE the 3D matrix of pathlosses is calculated, and the UE is attached to the home one (STBs) or the best one (other UEs) unless the best one is full, then it drops back to second best etc. UEs are attached in priority order, For every UE, SINR is calculated, required resource blocks is calculated and the femtocell loadings are tracked, max load on any femto = 90%, An abstraction of the scheduler assumes that interference can be mitigated from the two nearest neighbours, until the load-demand product exceeds a threshold

12 Model flow Input a probability of a broadband line having a femtocell Randomly place cells Put one of each type of UE around each cell Lowest pathloss is wanted femtocell. Calculate wanted signal Calculate pathloss to every femtocell and sort Start with highest priority UE Calculate unwanted signal by summing signals from other cells Calculate wanted / unwanted and look up spectral efficiency Calculate resource blocks taken by UE and keep track of loading Increment outage count for that UE type When loading > max then UE cannot get what it needs

13 Model flow Input a probability of a broadband line having a femtocell Randomly place cells Put one of each type of UE around each cell Lowest pathloss is wanted femtocell. Calculate wanted signal Calculate pathloss to every femtocell and sort Start with highest priority UE Calculate unwanted signal by summing signals from other cells Calculate wanted / unwanted and look up spectral efficiency Calculate resource blocks taken by UE and keep track of loading Repeated for other UEs Increment outage count for that UE type When loading > max then UE cannot get what it needs

14 Model flow Repeated for other probabilities of deployment to get required density Input a probability of a broadband line having a femtocell Randomly place cells Put one of each type of UE around each cell Lowest pathloss is wanted femtocell. Calculate wanted signal Calculate pathloss to every femtocell and sort Start with highest priority UE Calculate unwanted signal by summing signals from other cells Calculate wanted / unwanted and look up spectral efficiency Calculate resource blocks taken by UE and keep track of loading Repeated for other UEs Increment outage count for that UE type When loading > max then UE cannot get what it needs

15 System assumptions Parameter Value Comments Femtocell transmit power 13dBm per transmit port (maximum) Femtocell transmit power 3dBm per transmit port No UEs in connected mode (minimum) Frequency 2600MHz FDD UE receiver noise figure 7dB Bandwidth 15MHz Symbols per second per femtocell 12.6 Million This is 900 sub-carriers * 2000 slots per second * 7 symbols per slot Transmit overheads 3/14 This is for reference and control signals Maximum femtocell load 90% The percentage of resource blocks allocated / number available Minimum distance a UE can be from any femtocell 1m This limit is tighter than the use-case UEs and is imposed by the path-loss models Wall penetration loss (per external wall) 12dB mean, 2dB STD. Minimum loss capped at 2dB. This is for brick or block built housing. Across all building types, the mean is still 12dB but the STD climbs to 9dB. Indoor path-loss model ITU-R P dB 5dB correction added as a result of field measurements, caused by femtocell being low down, 500mm from floor. Outdoor path-loss model ITU-R P Includes LoS and non-los components with 20m transition region Minimum required SNR or SNIR -4dB For a UE to be able to attach (based on 1/12 QPSK)

16 Example results - outage probability No STB With STB at 10Mbit/s With STB at 20Mbit/s

17 Capacity estimate The spread of bit-rates across femtocells is commercially sensitive, Below are some broad results that we can share: Density limit /km 2 (femtocells) Total system capacity / Tbit/s 300 Max less 30% 900 Max (> 250) 1500 Max less 10% 2300 Max less 30% No limit A tiny bit less again So, from the point of view of outages, and of maximising system capacity, a density of about 1000 per km 2 is the optimum

18 Some guidelines from the modelling The maximum density is determined by the heaviest user, which in this case is the set-top box, STBs should be discouraged from being connected using LTE femtocells A density of around 1000 per km 2 may be a suitable target density It allows a good user experience with a low outage probability It gives the maximum system capacity

19 The question how well answered? Can we deploy LTE femtocells at a density that is high enough so we can send one to every BT broadband customer: - almost while supporting the applications in the business case Yes except for STBs and give inside-to-outside coverage that is comparable to WiFi Yes and give an improved experience with voice and data For voice yes, for data maybe for further work and give a reduction in customer complaints (propensity to call) Remains to be seen and cost the consumer no more, and cost BT as little as possible Ongoing work

20 The question how well answered? What is the maximum density of femtocells that: will support the applications and give adequate inside-to-outside coverage and give an improved user experience About 1000 per km 2

21 A bit about the MATLAB implementation Probability based (no time functions) Although we are now developing this, a per-tti simulation LTE toolbox not used in current model Biggest challenges Handling and sorting large files Input files are.txt with around 6 million lines Fastest way of sorting into histogram is using integer division and direct allocation to kilometre squares Working through large matrix calculations Up to 3000 * 3000 * 5 radio path-losses and summing interference Would do better by implementing a nearest neighbour function using tree structure, and then ignoring path-losses that are above a threshold

22 Example code for sorting broadband lines into tiles load 'bblocsx.txt' load 'bblocsy.txt' num = size(bblocsx); for i = 1:num xint(i) = uint32(round(bblocsx(i))); yint(i) = uint32(round(bblocsy(i))); end a = max(xint); b = max(yint); sq = zeros(a, b); % m = 1; % for i = 1:a % for j = 1:b % for k = 1:num % if xint(k) == a && yint(k) == b % sq(m) = sq(m) + 1; % end % end % m = m + 1; % end % end [a, b] = size(sq); sqr = reshape(sq, a * b, 1) Not efficient! for i = 1:num if xint(i)~= 0 && yint(i) ~= 0 % If the square km has a location sq(xint(i), yint(i)) = sq(xint(i), yint(i)) + 1; end end Efficient! [h, values] = hist(sqr(sqr~=0), 100); c = cumsum(h); x = sum(h);

23 Many thanks for listening