Many of the promises of LTE are well publicized:
- common standard across the globe
- high spectral efficiency (16.32 bits/Hz per mobile)
- potential to provide up to 326 Mbit/s per user
- all IP to overcome circuit switched legacies
- ability to use re-farmed spectrum
All of these point to a much lower operating cost for providing data services. Compared to the current cost levels (estimated to be around $0.01/MByte) for HSPA+, LTE is expected to reach another order of magnitude in cost reduction ($0.001/MByte).
Even with this rosy picture, the upcoming surge in data usage seems to be daunting. The following figure is from Cisco’s public research:
This graph and a companion (showing the distribution of total traffic among geographical regions) show that by 2014, total monthly traffic in the US will reach 750 PB (750,000 TB).
Based on the most recent CTIA figures, there are about 250,000 base stations in the US. Let’s assume for a second the “unimaginable” and think that a brand new LTE base station is placed next to an existing other technology base station and LTE carries only 50% of the total traffic (due to lagging deployment of devices) by 2014. With these assumptions, we can figure out what the typical base station traffic would be in 2014. Following table is my number crunching that shows the severity of the situation.
Like any large scale network, wireless service provider networks have fairly long tails in their traffic distribution graphs. Simply using the 80/20 rule, we can safely predict that approximately 20% of base stations will be carrying 80% of all traffic. If we stretch the same argument a bit further (assuming self-similarity of traffic distribution patterns), it is possible to reach to a conclusion that top 2.5% of all base stations will carry close to 1/3 of all traffic.
Similar to the long-tail geographical distribution of traffic, a significant temporal lumpiness also exist. In other words, traffic is not uniform, instead it increases/decreases with people’s sleeping, working, driving, shopping, etc. habits. That’s why instead of looking at the average throughput, we must consider the busiest hour in a given month. Typically I use a factor of 0.5% of the total monthly traffic in the busiest-hour for network design. This is a value I used in the past, even though it might be a bit conservative, it has served me well. As you see from the table, using this Busy-Hour loading factor, we end up with a throughput figure of 20 Mbps for the unlikely scenario of all base stations being equally loaded. On the other hand, if we look at the load of top 20% of base stations (that is 50,000 based on our very “unimaginable” scenario), then we end up with a requirement of 80 Mbps. Certainly the top 6% or 2.5% base station groups are much more difficult nuts to crack.
This tells me that by 2014, LTE with a theoretical sector capacity of 48 Mbps (in 4×4 MIMO downlink with 20+20 MHz) needs multiple base stations and more importantly much more spectrum. Considering this limit there is no alternative but to offload a large portion of this traffic to other wireless access networks. So far industry has come up with two possible solutions:
- Using LTE femtocells
- Using WiFi
Femtocell is certainly a viable approach as long as service providers subsidize it and are ready to handle the operational challenges of deploying and supporting them. However, femtocell doesn’t solve the impact of bringing this traffic back to the operator’s core network. Even though the fundamental access network constraint is resolved, operator will still be facing a significant investment to continue to scale its core network while managing millions of femtocells. Therefore, the alternative of using WiFi seems to be a very viable approach. Especially considering the unlicensed nature of spectrum, well-established home networking technology, commodity hardware and more importantly the presence of a WiFi radio in all new smartphones, tablets and other consumer electronic devices planned for LTE, WiFi offloading is the natural path to resolve the capacity crunch LTE networks will experience in 2014 and beyond.
Solving the capacity crunch via WiFi will also open up new ways in retrieving, storing, displaying content that will not be possible in an efficient way with LTE femtocells (remember the end-to-end principle). In the next post, I will cover some of those methods and how drastically they will change content distribution.