by Doug Pukallus - Network Engineer for Telco Antennas.
One of my quests over the past few weeks has been to test the upper limits of the Telstra 4G LTE network. My interest was to get as close to the 151.2Mbps theoretical maximum (physical layer) as possible - here in QLD we are lucky enough to be provided 15MHz of Telstra 1800MHz spectrum so our theoretical max is slightly higher than our New South Welshman and Victorian counterparts (10MHz bandwidth = 100.8Mbps theoretical maximum). For an explanation on calculations see the bottom of the page.
As you might have guessed that sort of speed is completely off the table for any real world transmission - as a deployed LTE-A service will have it's bandwidth (total resource blocks) eaten up by channel overheads (such as PDCCH, PCFICH, PHICH, PBCH) which reduce peak physical layer speeds down to between 107-129Mbps (PDCCH can consume 1-3 symbols per resource element). The remaining resource blocks are then shared out between all the users on the cell tower, which means to achieve anywhere close to these figures we need to minimise the number of users on the tower.
To approach this theoretical limit our next step is to improve channel quality. A good metric is our RSRP and SINR figures obtained from the modem's AT command interface. The RSRP figure indicates the received signal strength in dBm, and SINR indicates signal to noise ratio. Maximising our received signal power (RSRP) and reducing noise (SINR) is important to allowing our modem to negotiate a higher order of modulation, changing from QPSK -> 16QAM -> 64QAM -> 64QAM + 2x2MIMO as our channel quality improves.
Once we've obtained 64QAM and engaged both radios the next step is to push our coding rate. Coding rate refers to the amount of redundancy added to our transmission at the physical layer and is given as a coefficient between 0-1. This redundancy is designed to protect our physical layer frames from errors - note the distinction here, coding rate only deals with physical layer errors, retransmissions may still occur at the upper levels. Our coding rate is increased by maximising RSRP and SINR - providing an ultra clear signal, allowing us to pack more data into every block.
To limit the impact of other users I conduct tests between 1am and 2am. This is not perfect, my hope here is that people generally have better things to do at 2am. Speedtest.net was used as it forms the generally accepted and comparable metric.
I modelled 1800MHz transmission based on ACMA data, including antenna type, azimuth, AGL, downtilt, etc. to pick ideal locations where I'm likely to get a nice clean signal. Testing at a few different areas produced RSRP readings between -40 to -60dBm. Little improvement in speed was gained by achieving RSRP above -59dBm, the majority of improvements thereafter were made by maximising SINR.
Results were highly variable due to minor tower loading (as some late night users jumped on the tower), but were consistently around 87Mbps when I had it to myself. Peak application layer speeds were around 88Mbps, with the highest being 88.25Mbps.
In conclusion 90Mbps appears to be pushing the limits on 15MHz - it took over an hour to push from 87.95Mbps to 88.25Mbps, with different locations tested. The main limitation appears to be pilot interference and noise, reducing our coding rate. An important point to make was that the tests performed with Oolka (www.speedtest.net) were not of sufficient duration, as speeds were continuing to climb as the modem conversed with the eNodeB.
Since this test was conducted, our fastest QLD speed achieved is now 102.7Mbps from a newly activated 4G tower in an isolated area. Faster speeds are achievable with an LTE category 4 device in areas with 20MHz of spectrum.
2x 17dBi 1800MHz Grid antennas - arranged in 45/135 degree slant polarisation
2x LMR400 N to FME 5m cables
2x TS-9 to FME patch leads
Telstra 4G 320U USB (LTE category 3)
Theoretical Maximum Calculations
Given 15MHz of bandwidth we have 75 Physical Resource Blocks (PRBs), each consisting of 12 subcarriers with 7 symbols per subcarrier assuming normal cyclic prefix. Each subcarrier/symbol combination is called a Resource Element and can be transmitted in a 0.5ms slot. We use 64QAM so we encode 6 bits per symbol, and 2x2MIMO so we double our data rate. Therefore it is simply a matter of multiplication:
75 x 12 x 7 x 2 x 6 x 2 = 151200 bits per millisecond
151200 x bits/ms x 1000 x ms/s = 151200000bps
151200000bps = 151.2Mbps
 Harri Holma and Antti Toskala, LTE for UMTS: OFDMA and SC-FDMA Based Radio Access (Chichester: John Wiley &
Sons, 2009), 214.