Small Cells and the Predictability Challenge
December 14, 2012
Anyone closely involved with small cells base station could not have failed to notice the contentious debate on backhaul capacity requirements. On the one hand, vendors of millimeter wave equipment advocate the need to support peak capacity figures, while on the other hand, vendors of other types of backhaul solution including point-to-multipoint systems point that the peak is rarely if ever achieved in practice.
I don’t intend to tackle the topic of backhaul capacity requirements here, but I use this example to highlight a fundamental operational aspect to network design that lurks underneath such a debate. This aspect pertains to predictability of performance. Wireless networks by definition provide varying level of performance, so it becomes essential to achieve a certain level of predictability. Predictability is essential for network planning. It allows operators to dimension their networks meaning determine the number of sites and their resource allocations. It also allows operators to make marketing claims to position their service and commit to service level agreements. Predictability also impacts financial planning as predictable performance allows operators to better estimate the required expenditure.
Backhaul networks can be designed to very high level of performance predictability. Radio access networks on the other hand are less predictable, but operators have means to ensure a certain level of confidence. This has been standard practice in wireless network operations. The performance of macro-cells, while it varies, it can be predicated within certain margins (for example, it’s commonly acceptable that the standard deviation of log-normal propagation loss distribution is 8 dB which allows planning macrocells within margins that operators became comfortable with). The voice traffic on macrocells can also be predicted as it has historically followed certain patterns. The peak to average traffic on macrocells varies within limited margins which allows for certain level of predictability.
The issue is more complex when it comes to data networks and small cells. Data traffic is not as predictable as voice and has no specific patterns as it can happen at any time during the day. Also, small cells cover small areas which mean greater variability of number of underlying users. The traffic these users generate varies across wider margins. Furthermore, as small cells are located in closer proximity to users, there is greater potential of achieving higher data rates than that on a macrocells, but this will depend very much on subscriber density at the small cell as well on adjacent co-channel cells and macro-cells. What everyone agrees on is that small cells would have higher peak-to-average traffic ratio than macrocells. This wider variability contributes to lowering the predictability of performance.
When it comes to small cells, operators will be required to re-evaluate some of the processes used in designing wireless networks which were developed for macro-cells. For the reasons I outline here, new tools and techniques needs to be developed and implemented to plan for small cell networks. Margins of confidence have to be expanded to account for greater variability in performance. This aspect is critical to enabling the business case for small cells, or in other words, for small cells to take off.