1. RADIO NETWORK PLANNING
9.5 Inter-Cell Interference Coordination (ICIC)
A spread-spectrum system (like CDMA or UMTS) can work under very negative SINR because of the large processing gain for low data rates; soft handoff also helps tremendously. LTE air-link cannot work under the same negative SINR conditions, and does not support soft handoff. The industry recognized these cell- edge challenges, and responded by creating ICIC. Essentially, ICIC reduces the co-channel interference cell-edge users experience from direct neighbor cells, by increasing the cell-edge SINR values.
Traffic Channel ICIC
Although the scheduler has many dimensions to work with, the frequency and power domains are the main areas traffic-channel ICICs work with [17], [18]-[20].
In the frequency domain, the scheduler has the freedom to allocate any RB frequencies within the channel bandwidth. Working in the frequency domain is easy for traffic channels because the UE can easily find scheduling information from PDCCH, thus RB frequencies can be dynamically allocated. ICIC can allocate different RB frequencies to cell-edge users in different cells to avoid or
minimize co-channel interference with direct neighbors. Furthermore, LTE defines a few mechanisms to measure or notify direct neighbors of the out-of-cell interference levels on each RB (HII, OI, RNTP) [21].
Working in the power domain is also relatively easy, as fast power control is performed on the UL anyway. Also, the DL power level on selected RBs can be changed, for example, power boost for cell-edge users. Figure 18 shows two examples of traffic channel ICIC algorithms. The figure on the left shows an example for frequency-domain ICIC: frequency allocations for cell-edge users are restricted to about 1/3 of total channel bandwidth in order to avoid co-channel interference with direct neighbors; cell-center users can use the full bandwidth. The figure on the right shows an example of ICIC that works on both frequency domain and power domain: cell-edge users use one-third of the bandwidth but higher power; cell center users use full bandwidth but lower power.
Figure 29
Figure 29: Schematic diagram of various traffic channel ICIC solutions, showing ICIC in frequency domain (left) and ICIC in frequency and power domains (right)
ICIC on PDCCH?
The existing traffic-channel ICIC algorithms will not directly work on PDCCH, because PDCCH has a very different channel structure and is much less flexible. First of all, the scheduler does not have the freedom of trying to avoid co-channel interference by dynamically restricting the PDCCH bandwidth as it does with traffic channel RBs. Secondly; there are no X2 messages that support ICIC on PDCCH, at least not in the current release.
CCE-based power boost is one way the scheduler can play in the power domain.
CCE aggregation level can be 1, 2, 4 or 8 (CCE-1, CCE-2, CCE-4 or CCE-8). The higher the aggregation level, the more robust it will be. However, high aggregation levels also use more PDCCH resources. Therefore, cell-center users will use CCE-1 or CCE-2; users located somewhere in the middle will use CCE-2 or CCE-4; cell-edge users will always use CCE-8.
CCE-based power boost can boost up the transmit power level on CCE-8, which can
potentially increase the signal level on CCEs for cell-edge users.
How effective is the CCE-based power boost? Cells in a network can be in one of the following three scenarios:
• COVERAGE-LIMITED ENVIRONMENT
: The cells are spaced very far apart from each other. Examples are rural and highway cells. Typically the signal levels near the cell edges are already very low; as a result, the out-of-cell interference levels are also very low. For coverage-limited environments, one can approximate using the formula below:
In this case, boosting the signal power enhances “S,” and thus improves SNR since thermal noise is just a constant. So CCE-based power boost is effective in a coverage- limited environment.
• INTERFERENCE-LIMITED ENVIRONMENT:
The cells are packed very close to each other. Examples are dense suburban, urban or
dense urban with small cells. Typically the cell-edge composite signal level is very high, but the out-of-cell interference level is also very high. As a result, the cell-edge SINR is still poor. For interference-limited environment, one can approximate using the formula below:
In this case, CCE-based power boost will not be effective, because when signal power is
boosted up, the out-of-cell interference level is also increased, and as a result the SIR is not improved. Generally, when cell-edge power level is already very high, boosting the power further will not help.
o This phenomenon is the so-called “cocktail party effect:” in a cocktail party with high noise level in the background, it does not improve audibility if everyone increases their voice level; it just creates a higher level of background noise. o Unfortunately, an interference-limited environment is the area where help is most
needed. Call drops happen most frequently in small cells, especially calls placed from fast-moving vehicles.
o
10 CELL EDGE THROUGHPUT
A number of factors impact the ultimate capacity offered by a cell site. Two critical factors are interference and network loading. These factors are inter-related: higher network loading, which is a measure of the number of active subscribers, results in greater interference.
Interference determines signal quality and the modulation scheme: more bits can be sent over the air with higher modulation schemes. Low interference is typically achieved close to the cell center where distance to interfering cells is largest while highest interference is present at the cell edge where signal from the serving cell is weakest. Therefore, data rates are not even throughout the cell and they vary from highest close to the cell center to lowest at the cell edge.
To improve the capacity of a cellular network, different frequencies are used on adjacent cell sites. This is called „frequency reuse factor.‟ However, since LTE is designed to provide broadband services, it uses a wide channel bandwidth: 5, 10 or 20 MHz are the most common channel bandwidths (e.g. MetroPCS deployed 5 MHz system, Verizon deployed 10 MHz system and Telia Sonera 20 MHz system). This implies that standards frequency reuse as in traditional cellular networks is not possible due to lack of spectrum (e.g. Verizon‟s 700 MHz spectrum is 2×11 MHz). So, LTE networks are essentially reuse 1 networks. At full loading, one can expect significant interference, especially at the cell edge. Dropping the modulation rate helps in mitigating interference at the cost of reduced capacity. To further improve LTE capacity, clever techniques in assigning sub-carriers of the OFDMA physical layer to users are deployed which I can address in a future posting. The figure below shows the distribution of signal quality in a small network of 19 three- sectored cells for different re-use plans. The right most plot is for a single frequency reuse and shows the lowest signal quality performance, while the plot in the middle and to the left are for three and nine channels, respectively, where we have better performance.
Figure 30
Interference also works against MIMO spatial multiplexing which requires good signal quality to operate. Hence, it is likely that in large percentage of the cell coverage area, MIMO-SM is not operational due to interference, which results in lower capacity. Consider that peak LTE rates include a MIMO-SM capacity gain factor of 2, so the downlink LTE throughput for a 20 MHz channel is about 150 Mbps, or a spectral
efficiency (SE) of 7.5 b/s/Hz. Losing MIMO-SM halves the spectral efficiency down to 3.75 b/s/Hz.
The table below shows the average capacity for different LTE channelization‟s and corresponding spectral efficiency. Smaller channels have provide lower spectral efficiency that larger once because of control signaling overhead and loss of scheduling gain.
Channel Bandwidth
1.4 MHz 3 MHz 5 MHz 10 MHz 15 MHz 20 MHz
SE Relative to 10 MHz
62%
83%
99%
100%
100%
103%
Average Capacity (Mbps)
1.56
4.48
8.91
18
27
37.08
Average UL Capacity
(Mbps)
0.8
2.1
3.6
8
11
16
To conclude, single channel LTE networks will offer an improvement in spectral
efficiency over current 3G networks, but those expecting 150 Mbps download speeds will
be disappointed. Network operators cannot plan their network capacity based on peak
rates, but they will do so based on average rates which will be on the order of 1.4 – 1.8
b/s/Hz in the downlink.
11 VOIP CAPACITY CALCULATION
LTE supports voice over IP (Internet Protocol) (VoIP) to provide voice services. A simplified analysis is carried out next to approximately estimate the achievable VoIP capacity per cell in case of 10 MHz downlink channel bandwidth and 10 MHz uplink channel bandwidth. Refer to [2] for a comprehensive simulation-based analysis of VoIP capacity. We will carry out a simplified analysis below to estimate the VoIP capacity under a given set of assumptions. Refinement of assumptions and suitable modifications to calculations would lead to a more accurate prediction of VoIP capacity.
Assume that full-rate 12.2 kbps Adaptive Multi Rate (AMR) speech codec is used. Every 20 ms, AMR speech codec generates (12.2 kbps * 20 ms= 244 bits) during the “speech on” interval (i.e., the user is indeed talking and not just listening during such interval). These bits are placed in an RTP/UDP/IP packet with about 3 bytes (=24 bits) of overhead. IP header compression is assumed to be active. The VoIP packet entering the air interface protocol stack would contain about (244 speech bits + 24 IP-related header bits = 268) bits. The VoIP packet passes through these layers of the air interface protocol stack- PDCP, RLC, MAC, and PHY. Let‟s add 4 bytes (=32 bits) to account for headers added by PDCP (1 byte for short sequence number), RLC (1 byte for Unacknowledged Mode operation with a 5-bit sequence number), and MAC (2 bytes) layers, leading to the “target” payload of (268+32=300) bits entering the PHY layer from the MAC layer.
Now, let‟s calculate how many Physical Resource Blocks (PRBs) are needed to carry the target payload of 300 bits. According to Table A.3-1 of [36.104], 1 PRB can carry the payload of 104 bits when the modulation scheme is QPSK and the coding rate is (1/3). This payload is from the MAC layer to the PHY layer. Three PRBs would then be able to carry (104 bits per PRB *3 PRBs =312 bits), which would be adequate for the target payload of 300 bits. If a user‟s channel conditions allow the modulation scheme of 16-QAM and the coding rate of (3/4), 1 PRB can carry 408 bits, which would suffice for the target payload of 300 bits (see Table A.4-1 of [36.104]). When users are distributed across the cell, some would have good channel conditions and can support (16-QAM, coding rate=¾); others may have bad channels conditions and would require more robust (QPSK, coding rate=1/3). If 50% of users are able to use (16-QAM, coding rate=¾) and 50% of users need (QPSK, coding rate=1/3), the average number of PRBs consumed by a typical VoIP user in a cell would be (0.50*3 PRBs + 0.50*1 PRB = 2 PRBs). In 1 ms sub frame, there are 50 PRBs, allowing (50 PRBs/2 PRBs per user = 25 users). Since the AMR speech codec generates a new speech frame every 20 ms, during a span of 20 ms, we can have 20 sub frames carrying VoIP packets for (20 sub frames * 25 users per sub frame= 500) users. These calculations assume that every single packet with a specific modulation scheme and certain amount of coding is received without any errors all the time. However, in practice, some packets would be lost, requiring HARQ retransmission. If we need one (additional) retransmission on average, PRBS would need to be allocated to a given VoIP user twice per 20 ms interval instead of just once per 20 ms interval. Since a VoIP users is now consuming twice as many PRBs during the 20 ms interval, the number of VoIP users would be reduced by half (i.e., 500/2= 250). In summary, for the assumptions made here, the VoIP capacity in LTE is 250 in
case of 10 MHz channel bandwidth. Comprehensive simulation-based analysis indicates that 123 VoIP users can be supported in 5 MHz bandwidth [2], implying (123*2= 246) users can be supported in 10 MHz channel bandwidth.
The VoIP capacity estimate calculated above can be adjusted by modifying assumptions and making suitable adjustments to the calculations. For example, instead of using just two combinations of modulation scheme and coding rate, multiple combinations can be used to estimate the number of PRBs required by an average user in a cell. The overall approach outlined above can still be used for an approximate VoIP capacity estimate. Several factors would increase the VoIP capacity estimated above. If many users can
work with reduced degree of channel coding, capacity would be higher. We did not use 64-QAM in the analysis above, because only UE Category 5 can support such scheme in the uplink, and we may not see such UEs for quite some time. Use of antenna techniques would also increase the capacity. Consideration of the voice activity factor would also increase capacity because we do not need to send hundreds of speech bits during the silence interval. Some factors would decrease the VoIP capacity estimated above. If many users in the cell need more redundancy than that provided by (1/3) coding, we would need more PRBs per user, reducing the capacity. If semi-persistent scheduling is not used, higher control channel overhead would decrease the achievable VoIP capacity.
12 REFERENCES
[1]
3GPP, 36.104V8.7.0.[2] 3GPP, R1-072570, “Performance Evaluation Checkpoint: VoIP Summary.” [3] LTE in Bullets
[4] 3GPP TR-25.814- V7.1.0, “Physical Layer Aspects for E-UTRA,” ANNEX A: Simulation Scenarios, page 116.
[5] 3GPP TS-36.331- V10.2.0, “Radio Resource Control Protocol Specification,” Section 5.5.4.4: Event A3 (Neighbor becomes offset better than PCell), page 84.
[6] 3GPP TS-36.300- V10.4.0, “Overall Description,” Section 10.1.2, Mobility Management in ECM-CONNECTED, subsection 10.1.2.1.1, C-plane handling, page 62.
[6] 3GPP TS 36.133- V10.3.0, “Requirements for support of Radio Resource Management,” Section 7.6 Radio Link Monitoring, page 45.
[7] 3GPP TS-36.300 -V10.4.0, “Overall Description,” Section 10.12.6 Radio Link Failure, page 72.