It’s been a while since I wrote my last “random” post, so I thought it was about time to put up another. After all, sometimes there are just some observations that don’t quite warrant their own posting but are still worth getting online in some way.
Mobile Carriers Fight to the Bottom
The past half-a-year has seen a big competitive fight to the bottom in the Australian mobile phone carrier space. This seems to be something that has been initiated by Vodafone, initially offering half and even quarter-price starter kits with double (and even triple) data allowances on an already generous offering. As a result, other MVNOs have been following suit to various extent, including Catch Connect (a relatively young Optus-subsidiary that takes over the market space that Virgin Mobile used to occupy), Belong, Aldi Mobile, Kogan Mobile, amaysim, Lebara and more. Many of these are now offering discounted introductory pre-paid services which can be picked up for even as low as $0.99 for 40Gb in the first month as with the present Kogan Mobile offer. In some cases, SIMs purchased through referral shopping sites result in rebates greater than the cost of the SIM itself.
As a result, if you’re like me and don’t have any fixed line service and are facing waiting until at least April-July next year to receive HFC NBN, I suppose this is a godsend. Mobile data in Australia has traditionally been costly, with “unlimited” plans only just starting to crop-up, although with throttled speed and high cost. As a result, as a consumer it actually pays to hop from carrier to carrier, taking advantage of the lowest cost offerings in every month which could even end up saving money compared to a traditional fixed broadband connection.
The downside is that you’re likely needing to maintain a “real” number with a limited plan just for communication and “bootstrapping” the activation of new SIMs. It’s best to have a dedicated phone to operate a data-only SIM, or at least a dual-SIM phone otherwise. You also need to jump through “new customer” activation hoops every month, budget usage accordingly, plan ahead on the calendar and keep an eye out for what’s “cheap” for the next month keeping in mind when the offer must be activated by. In activating a new SIM, you realise that the card itself will be wasted after the month and that a number is going to be “tied up” for a while, although it will probably be on the receiving end of spam calls, so it’s best just so set a “call-forward-all” to voicemail to maintain your data sessions intact.
Knowing that it costs around $1 to send an envelope across Australia, I really wonder how they do this. I suspect Vodafone and their resellers especially are doing this for a numbers game – they might be looking to show good figures for customer sign-ups. But the truth is that customers are likely not going to stick for a number of reasons – so this “loss-leading” introductory strategy is likely to backfire especially when people like me can continuously take advantage of them. To increase the “stickiness”, it seems that Kogan have pushed up their data allowances from time to time and even gave away a free 3Gb data block to celebrate their anniversary. Other carriers are discounting 90 day plans instead, to lock users in for longer, although 365 day plans which used to be on sale have gone off promotion because they could result in users being locked into low prices for too long. It’s a fine balance, but I’m not sure it’s going to work out.
The Problem with Vodafone
In the past, I had analysed my connection speed through Vodafone and Kogan Mobile (a Vodafone reseller using the same APN and NAT infrastructure) to arrive at the conclusion that LTE is a poor substitute for a proper fixed connection. Needless to say, that was back in July and now with the addition of a number of “straws” onto the proverbial camel’s back, things have only gotten worse.
In October, I returned to Kogan Mobile and left my Fingbox to monitor the speed. Generally, speeds have declined somewhat – with some minimum speeds shown in the 6Mbit/s ballpark territory. The best speed is a paltry 16Mbit/s, but that is still better than some ADSL2+ connections.
But this doesn’t tell the entire truth. I wasn’t going to waste my quota on speed testing, so I turned it off, but then in the evenings when I wanted to get some work done, I found that connectivity was almost non-existent and slow as a crawl.
Notice how the download rate is much much lower than normal but the upload rate is just fine? This is a good sign that congestion is the problem. It was connected with LTE+ indicated on the screen … but this is hardly LTE-grade performance.
We will look into the details a little more in the following parts to try and explain why, but this is reminiscent of the 3G era when Optus was signing up MVNOs at breakneck pace – the network couldn’t handle the load and I ended up not using Optus for data at all in this area because it just couldn’t handle it. It seems that Vodafone’s limited spectrum and equipment capabilities along with an aggressive sales tactic might well come to haunt them with performance problems in certain areas.
Hello Catch? What’s the Catch?
In fact, this was only made extremely clear because the previous SIM I used was Catch Connect in the last 30-day cycle. With Catch, running on Optus, I experienced speeds much unlike that of Vodafone:
Of course, the speeds did vary – but when we’re talking about the slowest data point being about as fast as Vodafone’s best – this is not something to be snuffed at, although upload seemingly got progressively poorer as we went along. However, running Speedtest.net app got me results even higher – up to 98Mbit/s download which is truly amazing for a “lower-cost” option.
Of course, Catch being in their infancy had issues with their billing system – sending SMSes and e-mails about usage way too late (two days after expiry in my case). But I could forgive them for that, because the performance was so much better.
Delving into LTE Technical Details
To try and understand the disparity between the throughput, I tried to compare some of the technical details in regards to the LTE signal – frequency, bandwidth, spatial streams, tower distribution, customer loads, interference, etc. To help me with this, it helps that my “modem phone” is a Xiaomi Redmi Note 4X Qualcomm Edition, which has been rooted so that it can run Network Signal Guru.
Base Station Distribution
The first thing I wondered was whether there was anything different about the distribution of base stations. While it’s common for co-location agreements to be had on certain sites, the truth is that each carrier is responsible for their network planning and installing bases where they can.
Thanks to the ACMA Site Location Map, we can filter by the carrier to see what the distribution of bases looks like.
On the whole, the sites and their distribution are roughly similar between carriers. That was not a wholly unexpected finding as I am in the Sydney Metropolitan area, so demand for coverage will be there. Oddly, Telstra has the fewest bases with Optus having the most.
But the downside is that the base station coverage is somewhat sparse on the northern side of the map, which is an area where a lot of new housing developments that also do not have an NBN connection or anything else to take the load off. As a result, bandwidth demand is probably fairly high with very few bases to carry the data and poorer signal levels due to longer paths between the base and the user resulting in even lower airtime efficiency. So, unfortunately, my area is probably a bad case for congestion.
Bands and Carriers
In Australia, the 4G LTE bands in use are:
- B01 (2100Mhz)
- B03 (1800Mhz)
- B05 (850Mhz)
- B07 (2600Mhz)
- B08 (900Mhz)
- B28 (700Mhz)
- B40 (2300Mhz-TDD)
My Xiaomi Redmi Note 4X supports all of the bands with the exception of Band 8 and 28. Unfortunately, this means slightly sub-optimal speeds and coverage for Telstra which uses both and Optus which uses 28 additionally. However, that still leaves plenty of bands available for use.
Using my phone, I have observed the following frequencies in use in my area:
- B03 – EARFCN 1275 (1812.5Mhz) with 15Mhz bandwidth
- B07 – EARFCN 2950 (2640Mhz) with unknown bandwidth
- B07 – EARFCN 3148 (2659.8Mhz) with unknown bandwidth
- B03 – EARFCN 1725 (1857.5Mhz) with 15Mhz bandwidth
- B07 – EARFCN 3550 (2680Mhz) with 20Mhz bandwidth
- B40 – EARFCN 38770 (2312Mhz) with 20Mhz bandwidth
- B40 – EARFCN 38968 (2331.8Mhz) with 20Mhz bandwidth
- B40 – EARFCN 39166 (2351.6Mhz) with 20Mhz bandwidth
- B01 – EARFCN 75 (2117.5Mhz) with 15Mhz bandwidth
- B03 – EARFCN 1406 (1825.6Mhz) with 10Mhz bandwidth
- B03 – EARFCN 1550 (1840Mhz) with 20Mhz bandwidth
- B05 – EARFCN 2460 (875Mhz) with 10Mhz bandwidth
As I don’t use Telstra unless it is an absolute emergency, I didn’t observe many frequencies. Band 7 stations are normally used in high-density areas to offload high loads – you’ll find this near many train stations for example. With their arrangement, Band 7 carrier aggregation is very much a possibility that I have observed with my devices. Telstra also has a lot of Band 28 stations which I haven’t been able to observe.
Optus has a big distinct advantage – Band 40 acquired from Vividwireless which is a TDD network optimised for data throughput has a lot of available bandwidth. Each carrier is 20Mhz wide as most LTE modems cannot handle wider carriers, but as there is frequency diversity, many modems can use carrier aggregation to use a number of carriers at the same time. It was surprising to see that the Catch Connect SIM was permitted to connect on Band 40 to gain such good speeds and carrier aggregation was achievable in real life.
Unfortunately, Vodafone seems to be in a tough spot with relatively low bandwidths. Carrier aggregation on Band 3 seems to work, but otherwise no aggregation across bands seems to be permitted. From my connection data, it seems that the signal strength itself is good, but it seems to have been because of a large number of bases causing interference with each other. In the uplink direction, the phone is having to use almost the maximum power to send a return signal, so it seems that the sparse density of bases isn’t helping.
Carrier aggregation was possible, but things are even potentially worse – did you notice the MIMO settings?
It seems that Optus and Telstra both use base stations with 4-stream configurations. In the case of Vodafone, this is either 2-stream or 1-stream. For users with “mainstream” or lower end phones which only support two stream operation, four-stream support may seem like overkill, but having more MIMO streams is a way of supporting higher throughput with the same bandwidth and the fact there are still 1-stream eNBs on Vodafone’s network seems to be a sub-optimal choice.
Carrier aggregation is a way to try and make the most of base stations on different frequencies, and sometimes across different bands. Whether this is permitted depends on the base station set-up and the radio you’re using to access the network. Higher-end phones typically have more capable modems which will be able to aggregate more/wider carriers within the same band and across different bands.
One difference that can be seen is in the advertised AMBR (aggregate maximum bit-rate). On Telstra, this is 2000/2000 Mbit/s. On Optus, it is less at 1000/512 Mbit/s. On Vodafone, it is set to 536/208 Mbit/s, the lowest of the three. Of course, the realistic throughput achievable depends on real-life circumstances.
Trying a Solution
What is the solution? There really isn’t a real solution short of changing carriers. But there could be a way to help the situation. The first thing I thought of is to change the connection mode – changing over to 3G (+) results in …
… surprisingly, faster speeds than my LTE connection that night (<1Mbit/s). But of course, this isn’t very fast and 3G would limit the throughput even at non-congested periods to a lot less than LTE is capable of.
Instead, I decided to use the band-locking features of the software to disable bands to see how congestion affects different bands. Going over to Band 5 resulted in relatively slow and unstable speeds.
Trying the next band, Band 1 resulted in much better performance.
The throughput on Speedtest.net app also improved quite a bit at the most congested time.
Why is this the case? Well, each consumer mobile phone normally attempts to connect to the most suitable band. My presumption is that the phones aren’t so smart – they see B03 as the best option as there is two carriers to be aggregated for 30Mhz of spectrum. This would have the highest speed, assuming it was not congested. Because of the greedy nature of the modem, it just ends up making the congestion worse.
The 2100Mhz band is probably less used as its propagation is worse than the 1800Mhz band and much worse than the 850Mhz band, so it might “organically” have less users due to the problem of propagation. Band support of user equipment also factors into play. While locking to B01 reduces peak throughput, it avoids points in the day where throughput is <1Mbit/s and reduces congestion on B03 where everyone else is. Good for me, good for the others.
But alas, it looks like next month, I will be going with Catch again and getting slightly less quota instead. After all, the B40 base is normally only used by the Optus Home Wireless Broadband/Vividwireless products, so congestion doesn’t seem to be a big issue for now. The ~98Mbit/s speed is probably their rate limiting – I suspect the Snapdragon X9 could do better if the network didn’t limit it – in theory, 300Mbit/s downlink would be achievable using 2-streams per 20Mhz carrier x 2 carriers aggregated in ideal case.
Suspicious CAT5e Cable
Getting a few cables from eBay from random sellers is always a tricky proposition and this week, a very suspicious CAT5e cable arrived. The jacket was a little smaller than expected but squeezing it showed it was mostly empty inside. Uh oh.
Do you spot the problem? Well, this is not a proper UTP cable. There’s no twisting between pairs for one … another is the wires are too thin. There’s no way this cable works for gigabit, but it could work in a flaky way for 100Mbit/s … its functionally like CAT3 although with the full four-pair configuration.
The colours are also non-standard, so this is not kosher.
Just when you can’t imagine it getting any worse – the connector that was crimped had been cracked through the side, and they just “filled” it in with rubber and sold it anyway. This shouldn’t fly.
Just like many suspicious cables, a slight bend results in the plastic going “white”. This might take the cake as the worst network cable I’ve held in my hands.
Even the printing needs help – it says “NETOWORK PATCH CAT. 5E 24AWG 4PRETL TIA/EIA-568B 235M”. Missing a space and not even spelling the word network correctly, that’s just poor. It might as well say NOTWORK cable.
Unboxing Raspberry Pi 3 Model B+
I’ve always been a fan of the Raspberry Pi, being a beneficiary of their convenient form factor which helps my network do amazing things everyday. As a result, I’ve amassed a lot of Raspberry Pi boards over the years. While there’s been a few refreshes in the interim, the last board I had was the Raspberry Pi 3 Model B.
Since then, the Raspberry Pi 3 Model B+ was launched with an even better CPU, 5Ghz Wi-Fi, Bluetooth 4.2 with BLE support and PoE hat support amongst other changes. As I had some funding at my work to buy some supplies, I thought a Pi3B+ would be a worthy purchase from element14.
Unlike previous boxed Raspberry Pis, this one comes in a box that is “sealed” on the side to be opened by peeling back the flap. It wasn’t quite closed in good alignment, so it looks a little less neat.
Once flipping open the box, we are presented with the board straight away. That’s pretty cool, especially the stamped logo on the RF shielding can. The downside is that it doesn’t seem packed in an ESD shielding bag – could this be dangerous for the board? I notice there are a few discussions online about some boards suddenly failing – packing the board without an ESD bag might save a few cents but it doesn’t seem like the right thing to do.
There is a quick-start leaflet and a separator cardboard within the box. No more thick booklets that we don’t need. Great!
The board features a CPU with an integrated heatspreader, which should help with throttling. Unfortunately, no heatsink seems to be included but would be a nice addition. The power management seems to have been revised, with an MXL7704 as the controller. There is a header for the PoE hat to connect to. The LAN7515 chip now carries the Microchip logo rather than SMSC logo reflecting an acquisition. Aside from that, it seems the Wi-Fi antenna is now a trace on the board rather than a chip. The RF components are now shielded behind a can, with their logo stamped into it. This particular board was Made in the UK, with the USB connectors coming from Global Connector Technology instead of TE Connectivity as in the past.
The board still uses Elpida RAM and has a polyfuse on the underside. Lots of test-points abound which are useful for testing. I suspect the technology from Proant AB may be involved in designing the PCB antenna.
Aside from this, the footprint is very much compatible from the first Raspberry Pi Model B+, which is nice. Hopefully, I’ll have some time in the future to get this going – but I’ve only finally got my Asus Tinkerboard running in a permanent role.
Well, it’s been a long time since my last random posting and it’s no real surprise. Time is always in short supply. Life will always be full of things to do and not everything gets to make it to the blog. But hopefully, I’ll have some more time over the holidays to put more posts up – I’ve got a backlog of interesting data that needs analysing.