When we think about air interface capacity, we usually want to know how many users can be supported on a particular channel, or what kind of throughput they can get. These are important questions, but they are often difficult to answer accurately.  Capacity estimates are often based on an “average subscriber”, but there’s really no such thing. Today, Joe may be downloading a large video, while Mary is checking her email. Tomorrow, he may be browsing the web while she updates her Facebook status. Data usage is, by its very nature, wildly unpredictable.

A slightly easier question we can ask instead is: are we running out of capacity? Rather than trying to guess what the users are doing, we can instead look at the actual usage of the air interface resources. If any of those are nearing 100%, then we’ve reached the capacity limit of the cell. Well, not necessarily.

PRB and TTI Limits

In order to send and receive data over the LTE air interface, the eNB needs two things: Physical Resource Blocks (PRBs) to hold the data, and Transmission Time Intervals (TTIs) to send the data. The number of PRBs is defined by the bandwidth of the channel: a 5 MHz channel has 25 PRBs, for example, while a 20 MHz channel has 100. The eNB has an infinite supply of time, but only 1 TTI per millisecond.

The number of PRBs and TTIs the eNB uses to send a packet to a user depends on a large number of factors, including the size of the packet, the quality of the radio channel, the delay requirements for the application, and so on. The eNB may use 10 PRBs to send the packet in 1 TTI, or 10 TTIs to send the packet in only 1 PRB, or some other combination.

So it’s pretty straight-forward, then. If the eNB runs out of PRBs, or if it runs out of TTIs, and there’s still data in the buffers, then we have run out of capacity. Any data that didn’t get sent has to wait until resources are freed up, which means additional delay, which means lower throughput, which means lower user satisfaction. Well, not so fast.

The eNB vendors provide counters that track how many PRBs were used over the course of the reporting interval, and how many TTIs were needed to send that data. The calculation to determine the average PRB Usage and TTI Usage statistics for each channel is very simple:

  • PRB Usage = PRBs Used / PRBs Available
  • TTI Usage = TTIs Used / TTIs Available

So, for example, if the eNB reports that it used 40 PRBs on average when it sent data to the active users on a 10 MHz LTE channel, then the PRB Usage for that channel = 40/50 = 80%.

Suppose a particular cell shows a PRB Usage of 95% and a TTI Usage of 25% during the busy hour. That means the channel is nearly at maximum capacity, right? No, not really. All that means is that the eNB scheduler was dealing with bursts of traffic that used up 95% of the PRBs available in those particular TTIs. If any new data shows up in the buffers, it can be sent in another TTI, since only 25% were needed to ship out the previous data. The additional delay is generally negligible.

Suppose the usage numbers are flipped around: PRB Usage = 25%, and TTI Usage = 95%. Now the channel is full, right? Well, no. This pattern shows the eNB had a continuous stream of low speed data; there was always something to do, but the individual packets didn’t require many PRBs. If additional data shows up, the eNB can simply allocate more PRBs, since there are plenty left.

So if a high PRB Usage doesn’t mean the channel is full, and a high TTI Usage does mean that either, how do we know if we’re reaching the channel’s capacity limits? The answer, of course, is when both of these factors are high. A PRB Usage of 90% and a TTI Usage of 95%, for example, means that the eNB is nearly filling the pipe continuously, millisecond after millisecond. In this situation, the system is indeed reaching its capacity, and action should be taken to address it.

PDCCH and Scheduling Limits

So if the PRB Usage and TTI Usage are both low, it must follow that there are no air interface resource limitations, and everything is fine, right? Not so fast.

In order for the eNB to send data to the active users, it must first tell those users that the data is coming and provide instructions on where to look for it. This information is sent in the PDCCH, which itself has limited capacity; it can fill up quickly if the eNB needs to schedule a lot of users simultaneously. The vendors have various ways of measuring PDCCH occupancy, such as the number of symbols it uses (between 1 and 3), or the percentage of available Control Channel Elements (CCEs) that are used in the PDCCH. Either way, if the PDCCH is full, the eNB cannot send any more data until the next TTI, even if there are enough PRBs to hold the data.

Similarly, the eNB itself may be limited by how many scheduling decisions it can make in a TTI before it has to send what it has and move on. Currently, the major vendors allow only 5 users to be scheduled on the downlink in a single TTI, although this may increase over time. Even if there are still PRB, TTI and PDCCH resources available, the eNB must wait until the next TTI once it has assigned resources for 5 users. The end result is that, even if PRB and TTI usage are low, the air interface may still be at its capacity limit, if the PDCCH becomes full, or if the scheduler hits its limit.

The Cell is Full, Please Try Again Later

In summary, an LTE cell is at or near its traffic carrying capacity if any of the following conditions is true:

  1. PRB Usage and TTI Usage are both high (> 75%, say)
  2. PDCCH Occupancy is high (close to 3 symbols, or > 75% CCE usage, for example)
  3. The average number of scheduled users is close to 5 (or whatever the limit for that vendor is)

If you’re a capacity planner, you’re probably already looking at PRB Usage as a key capacity metric. Are you looking at these others as well?