The existing 1920s track circuit technology is simple and reliable. The train's wheels short the rails together, cutting off power to a relay at the end of the block, which sets signals behind the train to red. The classic system also detects broken rails and cars that are not part of trains. If someone leaves a freight car someplace, which happens, it's detected.

The classic system doesn't need GPS. The safety components ("vital" in signaling parlance) are very simple. The safety-related components are in trackside boxes along the right of way. There's lots of additional gear associated with dispatching and switching, and that's often centralized and computerized. But the basic train-presence detection and signal-setting is local.

There are later systems that still use track circuits, but provide more information to the driver and locomotive. These involve devices which sit between the rails and send signals to the train. Most of these are passive RFID-tag type devices. Some are wired into the signaling system.

The classic gear is old, but has a good track record.

Never buy trackside equipment from a place with a better climate than yours. - old railroad adage. The better railroad equipment suppliers used to be in upstate New York or Pittsburgh, where ice and snow are a normal part of life. Railroad equipment has to survive powerful snowplows, stuff dragging from trains, flamethrowers used to melt ice from switches, lightning strikes on the rails, machines which replace ties and ballast, and routine minor vandalism. There's a reason that gear is packaged in cast steel boxes.

1920s technology can provide enforcement. New York City subway and the Berlin S-Bahn both use old-school wayside trip cocks that raise next to red signals, and if a train passes a red, a mechanism on the side of the train strikes the raised trip cock, triggering a mandatory emergency brake application.

"The central principle of signal placement and engineering is that the control lengths of the signals be so designed such that when a train running at the highest possible speed under the worst weather conditions, with any semblance of functional brakes, is tripped, it will come to a halt before encountering the other train or obstruction that is causing the signal to be red."

Some signals have timers, for speed control. Go through a block too fast, and the signal ahead won't clear. Then the train stop will stop the train. This is common in downhill sections and ahead of curves.

Resetting a trip cock in NYC requires getting out of the cab and climbing down to the tracks. This is Not Fun.

Chicago has self-resetting trip cocks, which resulted in an accident in 2013.[1] A train parked in a service yard had a water leak into the control cables, which caused the train to move, slowly. The train hit a train stop, the brakes were applied, and the train stopped. But the trip cock reset, and the train started again. This happened five times at five different train stops. Workers noticed, but were not able to stop the train. This continued until the train hit an occupied but stopped train just outside the yard.

This is why you want dumb train control.

[1] https://www.ntsb.gov/investigations/AccidentReports/Reports/...

I'm not sure, though, that dumb train control is necessarily the answer. The Chicago system looks pretty dumb (though I know nothing about it), certainly no more complex than the New York system, but the fact that the CTA system self-reset seems to imply it cannot be safe without a human in the cab. This misfeature is a problem regardless how dumb or complex the signal system is.

Also, my hometown has one of the world's first automatic train operation systems (BART), and to my knowledge it hasn't had a revenue service malfunction leading to a crash since October 2 1972, less than a month after it opened. However, the train control system did lead to a good deal of litigation and project timeline overruns. I think the take-home lesson is that brand new technology is extraordinarily painful to debug and deploy; it's always going to be better to re-use train control that someone else debugged on their railway rather than re-inventing the wheel for your project.

I hope the US doen't get stuck with inventing something itself. It would certainly cost more. 20 or so non-European countries have started to use this system[1], including some very large ones.

[1] http://www.ertms.net/?page_id=55

There is a tradeoff of course, but it's not at all clear that the simplicity of track circuits outweighs increased utilisation of the track.

The enthusiasm for GPS-based systems comes from low-traffic railroads with a lot of track per train. Their maintenance cost for trackside equipment is high for the amount of traffic.

[1] https://www.theregister.co.uk/2016/08/09/london_underground_...

A strong argument in favour of computerisation, however, is that the worst delay caused to trains in service as a result of these types of failure was 30 minutes long – with the mean average delay time across all 700+ automated system failures being just five minutes.

That is amazing. Bear in mind that the Victoria Line runs over 30 trains per hour in the peak! I don't feel that "troublesome" is an appropriate description.

In the UK at least, a large number of railway routes really are constrained by the safety margins required for trains. In the late 90s, when the West Coast Main Line was upgraded, there was an abortive attempt to implement a modern moving-block signalling system – but it was predictably too difficult for such a complex line.

I don't doubt that you're right about the incentives being different on low-traffic lines though. It does demonstrate that there are various benefits to more automation that need to be traded off against the complexity, cost, and rust of failures.

London has at least some ATC on about half the lines, I don't think that's cautious.

A block is a discretization of the continuous track, and in track-space, your absolute "GPS" location isn't relevant -- just where you are on the track.

The on-vehicle location detector ("GPS") is essentially a cost-effective alternative to having infinitesimally small, fixed blocks.

The benefits are equivalent: the tradeoff is between whether you want to spend money on maintaining fixed trackside infrastructure, or more advanced equipment inside on-track vehicles.

The real solution is deploying moving block signals based on Communications-based train control (CBTC) which uses passive rfid balise's placed at known intervals and highly precise tachometers and odometers on the train. This allows any maintenance that needs to be conducted to be done offline, simply drive the train to the dedicated maintenance facility. With CBTC the only item needed to inspect on the right of way is radio system which can be done safely from the cat walk, ie. no need for track access or flagging.

This is what has been done for the Canarsie and Flushing Lines (except they also tie into the existing block signalling to support non cbtc compatible trains). CBTC allows for tighter headway's (more tph) and automatic train operation meaning reducing employe costs (trains drive themselves).

GPS is just a big distraction and should be avoided for mass transit and reserved basically for dark territory PTC.

The modern systems generally prevent this. See [1] for an exception -- the engineers forgot to install the limiting markers at the very end of the high speed line, allowing the train to overspeed into the normal speed line.

[1] https://en.wikipedia.org/wiki/Santiago_de_Compostela_derailm...

Why do things this way, instead of just having the switches themselves notify control when they change state? (Or do they already?) That by itself tells you almost all you need to know about where a train is.

Of course they use the rail as an electrical ground, though, which could be an issue.

But it is correct for you to say "to the next obstacle" rather than "to the train". The next obstacle could be a switch, or perhaps a line welded on for the signaling system. It could even be a broken rail.

The problem is that detecting that serves no purpose, and you need to detect past those joins for it to serve a use.

Understatement of the year. Even in the US the number of deaths per year averages around half a dozen; you're about ten times as likely to be killed by a bolt of lightning.

From a people-safety perspective, encouraging people to use trains by making them convenient will have an overwhelmingly better impact than trying to improve train safety.

Not that they are mutually exclusive. But if you have a million dollars, I bet you save more lives by putting out a TV ad saying "Thinking of driving? Why not take the train?" than by upgrading a stretch of track. Or by using it to bribe popsci.com into not writing articles about trains subheaded with words like "frustrating", "delays", and "accidents".

https://www.jefftk.com/p/make-buses-dangerous

If you want to see conventional, FRA-approved "buff strength" safety tested against "crash energy management" -- a broader concept including crumple zones -- watch this video [1]. The differences are striking.

But to reiterate: don't blame crash safety on lacking technology. Blame it on bad regulation.

[1] https://www.youtube.com/watch?v=NUpUJrk4QBE

Outdated regulation plays a part, particularly around the assumptions of where loads come from. As you mentioned, the buff crushing load assumes a longitudinal action only. Moreover, there has historically been minimal roll-over strength in cabins to protect pasengers. Change is disappointingly slow, however, and with asset lives of 50+ years due to the tough financial challenges facing particularly passenger rail, it takes a long time for new regulation to come into practice.