What I’d like to talk about
▪ The need for RRM (briefly),
▪ Some research:
A technique for choosing the best radio access technology
(RAT) from a set of options, using fuzzy logic and multiple-
attribute decision making (MADM)
Simulation results using video and FTP traffic over licensed and
unlicensed spectrum,
▪ Key findings and planned further work
2
Acknowledgment
The research reported in this presentation was conducted by the Speed-5G project that receives
funding from the European Commission H2020 programme under Grant Agreement N: 671705.
The European Commission has no responsibility for the content of this presentation.
The need for Radio Resource Management
▪ RRM will increase spectrum utility if spectrum and Radio Access
Technology (RAT) type can be flexibly matched with link
requirements, for example by:
Making most use of unlicensed spectrum,
Using high frequencies and low powers for short ranges.
3
RRM has to do the following:
▪ Work within the constraints of spectrum regulation,
▪ Decouple RAT from spectrum type:
I assume that all future auctioned licensed spectrum will be
technology neutral (as is most current licensed spectrum)
Unlicensed spectrum normally uses WiFi, but LTE is also
possible:
LAA uses licensed and unlicensed spectrum, where a licensed anchor is
used for signalling and high QoS data, and unlicensed carriers are used for
other data. LBT is used,
LTE-U has the control plane over licensed and all data over unlicensed and
does not use LBT and so is not used in Europe,
MuLTEfire operates LTE entirely in unlicensed spectrum (and uses LBT).
4
Software defined radio is a key enabler for RRM
RRM functions on the control plane and steers
the MAC towards RAT type and spectrum to use
for each session,
The KPI collector keeps track of KPIs per session,
The Spectrum Manager is the available portfolio
of spectrum with policies for its use.
RAT / spectrum selection: an RRM algorithm using
fuzzy logic and MADM
6
New application
to be established
Out of context
suitability levels
(fuzzy logic)
In context
suitability levels
(MADM)
Short term RAT attributes
(eg from reference
signals)
Context components
(eg power, velocity)
Medium and long term
RAT attributes and
MADM parameters
(eg cost, range, weights)
Decision: rank
RATs and choose
best one
Step 1
Step 2
Step 3
Design requirements
▪ Context-awareness needed to exploit the various context
components associated with:
Applications (e.g. QoS requirements)
UE (e.g. battery level and velocity)
Network (e.g. operator strategies and regulation rules),
▪ User-driven fast decisions (e.g. switch to traditional RATs out of
5G coverage),
▪ Network assistance to provide UEs with the relevant information
and offer flexibility to the network manager.
8
Inputs to RAT decision process
▪ X applications with various requirements
𝐴𝑥 (1<𝑥<𝑋)
▪ K available RATs 𝑅𝐴𝑇𝑘 (1<𝑘<𝐾)
▪ S subscription profiles 𝑃𝑠 (1<𝑠<𝑆)
▪ Various context components
9
Step 1: Estimate out of context suitability level 𝑆𝑘,𝑥𝑜𝑐 for
each RAT / application pair using fuzzy logic
10
FLCRAT
parameters
𝑆𝑘,𝑥𝑜𝑐
RAT parameters can be RSRP, time waiting in buffers, packet loss rate etc
and may require some pre-processing such as averaging. The FLC runs a set
of inference rules to combine multiple inputs of multiple levels, to a single output.
1
-20 -19 -16 -15 RSRP/dBm
Degree of membership
Low Medium High
𝑆𝐿𝑇𝐸,𝑣𝑖𝑑𝑒𝑜𝑜𝑐0.8 0.9
Degree of membership
Low High
1
Example input RAT parameter Example output
Step 2: Use Multi-Attribute Decision Making to include
context and also to include the 𝑆𝑘,𝑥𝑜𝑐 from step 1
(a) Construct decision matrix Dx whose elements 𝑑𝑘,𝑚𝑥 denotes the performance of
RATk in terms of the mth attribute:
11
𝐷𝑥 =
(b) Derive a normalised matrix 𝐷𝑥 whose elements 𝑑𝑘,𝑚𝑥 are determined as:
𝑑𝑘,𝑚𝑥 =
𝑑𝑘,𝑚𝑥 /max
𝑘𝑑𝑘,𝑚𝑥
min𝑘
𝑑𝑘,𝑚𝑥 /𝑑𝑘,𝑚
𝑥
for benefit attributes (eg QoS/QoE and range)
for cost attributes (eg cost and power)
RAT QoS/QoE Cost Power Range
RAT1 𝑆1,𝑥𝑜𝑐 Cost1 Power1 Range1
• • • • • • • • • •
RATk 𝑆𝑘 ,𝑥𝑜𝑐 Costk Powerk Rangek
• • • • • • • • • •
RATK 𝑆𝐾,𝑥𝑜𝑐 CostK PowerK RangeK
Step 3: Apply weights and choose best RAT
(the weights are adjusted according to operator policy)
(c) Determine the in-context suitability levels by multiplying by weights
12
= 𝐷𝑥 ∗ 𝑊𝑥𝑠
𝑊𝑥𝑠 =
(d) Select the kth RAT that maximises the in-context suitability levels
depending on x application and subscription level s
𝑘∗ 𝑥, 𝑠 = argmax𝑘
𝑆𝑘,𝑥𝑖𝑐,𝑠
𝑆1,𝑥𝑖𝑐 ,𝑠
• •
𝑆𝑘 ,𝑥𝑖𝑐 ,𝑠
• •
𝑆𝐾 ,𝑥𝑖𝑐 ,𝑠
𝑤𝑥 ,𝑄𝑜𝑠𝑠
𝑤𝑥 ,𝑐𝑜𝑠𝑡𝑠
𝑤𝑥 ,𝑝𝑜𝑤𝑒𝑟𝑠
𝑤𝑥 ,𝑟𝑎𝑛𝑔𝑒𝑠
Use-case for testing
▪ X = 2 applications:
Interactive video with 100ms maximum latency and max frame loss = 0.1
FTP as on-off process where each ‘on’ consumes a fraction ρ of the capacity of
the in-use RAT with loose QoS requirements
▪ K = 3 RATs
LTE: small cells operating in licensed band
WLAN: APs operating in unlicensed band
LBT: small cells sharing WLAN band on listen before talk
▪ S = 2 subscription profiles
Gold which is flat rate
Bronze which is limited credit
13
K = LTE, X = video
14
- RSRQ = reference signal received quality
- T_Sched = time packets wait in buffer before
being transmitted
K = WLAN, X = video
15
- SINR = signal to interference plus noise ratio
- T_ACK = time packets wait in buffer before
being successfully transmitted (ie acknowledged)
- Drop_R = rate of dropped packets due to overloaded
MAC or unsuccessful re-transmissions
K = LBT, X = video
16
- RSRQ = reference signal received quality
- T_Access = time packets wait in buffer before
accessing the channel
- NACK_R = the ratio of NACKs out of the HARQ-ACK
feedback values
Step 2, add the in-context attributes to form the Dx and
𝐷𝑥 matrices
17
RAT QoS/QoE Cost Power Range
LTE 𝑆𝐿𝑇𝐸 ,𝑣𝑖𝑑𝑒𝑜𝑜𝑐 High High Large
WLAN 𝑆𝑊𝐿𝐴𝑁 ,𝑣𝑖𝑑𝑒𝑜𝑜𝑐 Low Medium Small
LBT 𝑆𝐿𝐵𝑇 ,𝑣𝑖𝑑𝑒𝑜𝑜𝑐 Low High Small
RAT QoS/QoE Cost Power Range
LTE High High High Large WLAN High Low Medium Small
LBT High Low High Small
Dvideo =
DFTP =
Derive the normalised matrix 𝐷𝑥 whose elements 𝑑𝑘,𝑚𝑥 are determined as:
𝑑𝑘,𝑚𝑥 =
𝑑𝑘,𝑚𝑥 /max
𝑘𝑑𝑘,𝑚𝑥
min𝑘
𝑑𝑘,𝑚𝑥 /𝑑𝑘,𝑚
𝑥
for benefit attributes (eg QoS/QoE and range)
for cost attributes (eg cost and power)
Step 3: Apply weights and choose best RAT
18
= 𝐷𝑣𝑖𝑑𝑒𝑜 ∗ 𝑊𝑣𝑖𝑑𝑒𝑜𝑠
𝑊𝑣𝑖𝑑𝑒𝑜𝐺 =
𝑤𝑣𝑖𝑑𝑒𝑜 ,𝑄𝑜𝑠𝐺 = High
𝑤𝑣𝑖𝑑𝑒𝑜 ,𝑐𝑜𝑠𝑡𝐺 = High
𝑤𝑣𝑖𝑑𝑒𝑜 ,𝑝𝑜𝑤𝑒𝑟𝐺 = High
𝑤𝑣𝑖𝑑𝑒𝑜 ,𝑟𝑎𝑛𝑔𝑒𝐺 = High
𝑊𝑣𝑖𝑑𝑒𝑜𝐵 =
𝑤𝑣𝑖𝑑𝑒𝑜 ,𝑄𝑜𝑠𝐵 = High
𝑤𝑣𝑖𝑑𝑒𝑜 ,𝑐𝑜𝑠𝑡𝐵 = Low
𝑤𝑣𝑖𝑑𝑒𝑜 ,𝑝𝑜𝑤𝑒𝑟𝐵 = High
𝑤𝑣𝑖𝑑𝑒𝑜 ,𝑟𝑎𝑛𝑔𝑒𝐵 = High
𝑆𝐿𝑇𝐸 ,𝑣𝑖𝑑𝑒𝑜𝑖𝑐 ,𝑠
𝑆𝑊𝐿𝐴𝑁 ,𝑣𝑖𝑑𝑒𝑜𝑖𝑐 ,𝑠
𝑆𝐿𝐵𝑇 ,𝑣𝑖𝑑𝑒𝑜𝑖𝑐 ,𝑠
And another diagram like this for x = FTP that is
very similar.
Then finally the best k RAT for x application and
s subscription is
𝑘∗ 𝑥, 𝑠 = argmax𝑘
𝑆𝑘,𝑥𝑖𝑐,𝑠
Video test set-up
19
QoE parameters used were:
1. Peak signal to noise ratio (PSNR): measures similarity based on a frame-by-frame
comparison.
2. Structural Similarity (SSIM) index: focuses on the structural information loss, to
which our eyes are strongly sensitive.
Sequence was 2-min Big Buck Bunny sequence (16+ million views on Youtube).
It is encoded with H264 (Main Profile, L4) at 1080p @24fps.
Model is ultra-dense environment
20
1. Single macro-cell overlaid by various buildings.
2. A small-cell randomly deployed in each room on top of an AP.
3. Small-cells are dual-access i.e., share licensed bands with macro (co-channel)
and unlicensed bands with existing WLAN based on listen-before-talk (LBT).
Simulation assumptions
▪ Single cell scenario (with interference and pathlosses taken
from ultra-dense environment)
▪ Gold and Bronze video sessions with one FTP transfer initially
on WLAN.
▪ Fraction of used capacity by FTP (ρ) used to load the WLAN (so
FTP stays on WLAN)
▪ May choose LTE / WLAN (offload) / LBT (sharing) for video and
switch between them
21
Key findings and further planned work
▪ Gold user stayed on LTE the whole time,
▪ Bronze user: significant gains due to utilisation of unlicensed
band:
Offloading better at low WLAN loads, sharing better at high
WLAN loads,
Proposed approach switches efficiently between both
options, 12dB improvement in PSNR,
▪ Further planned work:
More RATs (e.g. mm-wave) and more applications (e.g.
adaptive video streaming and massive IoT),
Real-world trials.
26
Use of spectrum
▪ Spectrum is loosely categorised in terms of licensed, lightly licensed, or
unlicensed
Strictly though, it is the operator who is licensed to operate equipment that uses
the spectrum.
▪ The licence can be Spectrum Access (technology neutral) or Technology
Specific.
Examples of TS bands are the 1800MHz that is specified for 3G and LTE only,
and the low power GSM (LPGSM) band at 1900MHz that is specified for 2G
and LTE only
The trend is towards SA, and all future auctions are assumed to be for SA
spectrum
▪ The licence specifies limits for transmit power, channel widths, in-band and out-
of-band unwanted emissions, specifies TDD / FDD and can include additional
ETSI or 3GPP specifications
▪ Different equipment in the same system is licensed in different ways
In LTE for example, the base-stations operate on a licensed basis but the UEs
operate on an unlicensed basis because they are owned by individuals and
they move around
28
Use of spectrum
▪ The holder of a spectrum licence has sole use of that spectrum
It enables the holder to manage interference and offer controlled QoS to its
customers. It allows the highest transmit powers of all licence options and
hence has the longest range
It is also expensive, and is best reserved for high QoS high value services
▪ Lightly licensed equipment is a sensible option only for small planned
deployments
The regulator keeps a record of locations of BSs, and there is an annual fee for
every piece of equipment. It is shared spectrum so QoS guarantees are not
possible.
Medium powers are used so the range is medium (a few km). An example of its
use is TV Whitespace. It is not a good option for planned or mass unplanned
deployments like LTE or WiFi
▪ Unlicensed equipment can be used by anyone
Is free, is best effort, and needs a form of contention management that works
across equipment from different operators.
It is low power and hence works only over short ranges (eg indoors)
29