10158 stories
·
34 followers

On the Diplomacy AI

2 Shares

The latest AI development is: AI achieves human level in (blitz 5-minute-turn) full-communication anonymous online Diplomacy (paper). Why not?

I mean, aside from the obvious.

A take I saw multiple times was that AI labs, or at least Meta, were intentionally going for the scariest possible thing, which is why you create the torment nexus, or in this case teach the AI to play Diplomacy. If you had to pick a game to sound scary, you’d definitely pick Diplomacy.

The universal expectations for AI breakthroughs like this are:

  1. The particular breakthrough was not expected, and is scary. The techniques used worked better than we expected, which is scary.
  2. The details of the breakthrough involve someone figuring out why this particular problem configuration was easier to solve than you would expect relative to other problems and configurations, and thus makes it less scary.
  3. We find that those details matter a lot for success, and that close variants would not be so easy. Other times we will find that those details allowed those creating the new thing to skip non-trivial but highly doable steps, that they could go back and do if necessary.

That is all exactly what we find here.

The actual AI, as I understand it, is a combination of a language model and a strategic engine.

The strategic engine, as I evaluated it based on a sample game with six bots and a human, is mediocre at tactics and lousy at strategy. Humans are bad at tactics (and often strategy) in games and Diplomacy is no exception. Diplomacy’s tactics a good match for a AI. Anticipating other players proved harder. The whole thing feels like it is ‘missing a step.’

What Makes the AI Good?

Where does the AI’s advantage come from? From my reading, which comes largely from the sample game in this video, it comes from the particulars of the format, and not making some common and costly mistakes humans make. In particular:

  1. AI writes relatively long, detailed and explanatory communications to others.
  2. AI does not signal its intentions via failing to communicate with its victims.
  3. AI understands that the game ends after 1908 and modifies accordingly.
  4. AI keeps a close eye on strategic balance in order to maximize win percentage.
  5. AI uses its anonymity and one-shot nature to not retaliate after backstabs.
  6. AI knows what humans are like. Humans were not adjusted to bot behaviors.

When people say the AI ‘solved’ Diplomacy, it really really didn’t. What it did, which is still impressive, is get a handle on the basics of Diplomacy, in this particular context where bots cannot be identified and are in the minority, and in particular where message detail is sufficiently limited that it can use an LLM to be able to communicate with humans reasonably and not be identified.

If this program entered the world championships, with full length turns, I would not expect it to do well in its current form, although I would not be shocked if further efforts could fix this (or if they proved surprisingly tricky).

Interestingly, this AI is programmed not to mislead the player on purpose, although it will absolutely go back on its word if it feels like it. This is closer to correct than most players think but a huge weakness in key moments and is highly exploitable if someone knows this and is willing and able to ‘check in’ every turn.

The AI is thus heavily optimized for exactly the world in which it succeeded.

  1. Five minute turns limit human ability to think, plan and talk, whereas for a computer five minutes is an eternity. Longer time favors humans.
  2. Anonymity of bots prevents exploitation of their weaknesses if you can’t confidently identify who they are, and the time limit kept most players too busy to try and confidently figure this out. They also hadn’t had time to learn how the bots functioned and what to expect, even when they did ID them.
  3. One-shot nature of games allows players to ignore their reputations and changes the game theory, in ways that are not natural for humans.
  4. Limited time frame limits punishment for AI’s inability to think about longer term multi-polar dynamics, including psychological factors and game theoretically strange endgame decisions.
  5. Limited time frame means game ends abruptly in 1908 (game begins in 1901, each year is two movement turns, two retreats and a build) in a way that many players won’t properly backward chain for until rather late, and also a lot of players will psychologically be unable to ignore the longer term implications even though they are not scored. In the video I discuss, there is an abrupt ‘oh right game is going to end soon’ inflection point in 1907 by the human.
  6. Rank scoring plus ending after 1908 means it is right to backstab leaders and to do a kind of strange strategy where one is somewhat cooperating with players you are also somewhat fighting, and humans are really bad at this and in my experience they often get mad at you for even trying.

The Core Skill of Online Diplomacy is Talking a Lot

As the video’s narrator explains: The key to getting along with players in online Diplomacy is to be willing to talk to them in detail, and share your thoughts. Each player only has so much time and attention to devote to talking to six other players. Investing in someone is a sign you see a future with them, and letting them know how you are thinking helps them navigate the game overall and your future actions, and makes you a more attractive alliance partner.

Humans also have a strong natural tendency to talk a lot with those they want to ally with, and to be very curt with those they intend to attack or especially backstab (or that they recently attacked or backstabbed). This very much matches my experiences playing online. If a human suddenly starts sending much shorter messages or not talking to you at all, you should assume you are getting stabbed. If you do this to someone else, assume they expect a stabbing. Never take anyone for granted, including those you are about to stab.

This gives the AI a clear opportunity for big advantage. An AI can easily give complex and detailed answers to all six opponents at the same time, for the entire game, in a way a human cannot. That gives them a huge edge. Combine that with humans being relatively bad at Diplomacy tactics (and oh my, they’re quite bad), plus the bots being hidden and thus able to play for their best interests after being stabbed without everyone else knowing this and thus stabbing them, and the dynamics of what actually scores points in a blitz game being counter-intuitive, and the AI has some pretty big edges to exploit.

The five minute turns clearly work to the AI’s advantage. The AI essentially suffers not at all from the time pressure, whereas five minutes is very little time for a human to think. I expect AI performance to degrade relative to humans with longer negotiation periods.

Lessons From the Sample Game

The sample game is great, featuring the player written about here. If you are familiar with Diplomacy or otherwise want more color, I recommend watching the video.

The human player is Russia. He gets himself into big trouble early on by making two key mistakes. He gets out of that trouble because the AI is not good at anticipating certain decisions, a key backstab happens exactly when needed, the player wins a key coin flip decision, and he shifts his strategy into exploiting the tendencies of the bots.

The first big mistake he makes is not committing a third unit to the north. Everything about the situation and his strategy screams to put a third unit in the north, at least an army and ideally a fleet, because the south does not require an additional commitment or does the additional commitment open up opportunity. Instead, without a third northern unit, Russia has nowhere to expand for a long time.

The second big mistake was violating his DMZ agreement with Austria by moving into Galicia. He did this because the AI failed to respond to him during the turn in question, and he was worried this indicated he was about to get stabbed, despite the stab not making a ton of tactical sense. Breaking the agreement with Austria led to a war that was almost fatal (or at least probably did, there’s some chance Austria does it anyway), without any prospect of things going well for Russia at any point.

Against a human, would this play have been reasonable? That depends on how reliable an indicator is radio silence, and how likely a human would be to buy it as an excuse. Against an AI, it does not make sense. The AI has no reason to not talk at all in this spot, regardless of its intentions. So it is strange that it did not respond here, it seems like a rather painful bug.

The cavalry saves us. Italy stabs Austria, while France moves against England.

Here is a tactical snapshot. I hate France’s tactical play, both its actual plays and the communications with Russia that are based on its tactics, dating back to at least 1903. The move here to Irish Sea needs to be accompanied by a convoy of Picardy into London or Wales, fighting for Belgium here is silly. Italy does reasonable things. Austria being in Rumania and Ukraine is an existential threat, luckily Austria chooses a retreat here that makes little sense. Once you have Bulgaria against Turkey, you really don’t want to give it up. Austria also lost three or so distinct guessing games here on the same turn. Finally I would note that Italy is surprisingly willing to lose the Ionian Sea to pick up the Aegean, and that if I am Turkey here there is zero chance I am moving Ankara anywhere but Black Sea.

My sense is also that the AI ‘plays it safe’ and does what it thinks is ‘natural’ more often than is game theory optimal. This is confirmed by an author of the paper here, along with other similar observations. The AI assumes it can ‘get away with’ everything because on the internet no one knows you are a bot or what you are up to, and makes decisions accordingly. A huge edge if you get away with it. A huge weakness if you do not.

Then again, Diplomacy players are weird, myself included. There is almost always a tactical way to punish an aggressive ‘natural’ or ‘correct’ play if you are willing to get punished hard by other moves, such as if Germany were to try to sneak into Picardy (PIC) here. So any given decision could be one mixing up one’s play, so my evaluations are more based on the whole of the eight years of play by six players.

The turn above, Spring 1904, is about where Russia pivots from acting like it is playing a normal full game against humans to understanding it is playing an eight-year game for rank order against bots, and he starts asking ‘what would a bot do?’ Things turn around quite a bit after that. His only slip beyond that is at about 42:00 when he worries he will ‘annoy Austria’ in a way that shouldn’t (and didn’t) apply to a bot.

The big exploit of the bots is simple. A bot is not going to retaliate later in the game for a backstab earlier in the game, or at least will retaliate far less. As things shift into the endgame, taking whatever tactical advantages present themselves becomes more and more attractive as an option. Bots will sometimes talk about ‘throwing their centers’ to another player as retaliation, or otherwise punishing an attacker or backstabber, but you know it is mostly talk.

If you play Diplomacy using pure Causal Decision Theory without credible precommitments, and it is a one-shot fully anonymous game, that can work. When you are identifiable (or even worse if someone can see your source code, as they could in a lot of MIRI or other old-school LW thought experiments), you are going to have a bad time.

Diplomatic Decision Theory

The central decision theory question of Diplomacy is how one should respond when stabbed, and what this says about how one should act before one is stabbed.

Responses run the whole range from shrugging it off to devoting the rest of one’s life to revenge. There is a reason people say Diplomacy ruins friendships. Reasonable people max out at ‘spend the rest of the game ensuring you lose’ and being less inclined to trust you in future games, but a lot of what keeps human systems working is that you never know for sure how far things might go.

When deciding whether to attack someone, a key consideration is how they are likely to react. If they are going to go kamikaze on you, you need to ensure you can handle that. If they are going to mostly shrug it off, even let you use your newly strong position to drive a better bargain, then it is open season whenever you have a tactical opening, and then there is everything in between.

The correct solution in a fully one-shot anonymous game, if you can pull it off, is obviously to give people the impression you will strongly retaliate, then to not follow through on that under most circumstances. Humans, of course, have a hard time pulling this off.

Bots also have a hard time pulling this off in a credible way, for different reasons. The bots here mostly were free riders. Humans did not know what they were dealing with. So they gave bots an appropriately broad range of potential reactions. Then the bots got the benefits of not spending their resources on punishment. Once humans did know what they were dealing with, and adjusted, things wouldn’t go so well there. If there were a variety of bots competing at that point, bots would have a hell of a time trying to represent that they would actually retaliate ‘properly.’

Thus, the ‘irrational’ flaws in humans grant them a distinct advantage in the default case, where identity is broadly (partially, at least) known and behaviors have a chance to adjust to what information is available.

AIs so far have essentially ‘gotten away with’ using Causal Decision Theory in these spots, despite its extreme vulnerability to exploitation. This contrasts with many much ‘dumber’ AIs of the past, such as those for Civilization, which were hardcoded with extreme retaliation functions that solve these issues, albeit at what could be a steep price. I wonder what will happen here with, for example, self-driving cars. If AIs are going to be operating in the real world more and more, where similar situations arise, they are going to have to get a better decision theory, or things are going to go very badly for them and also for us.

In this sense, the Hard Problem of Diplomacy has not yet been touched.

Overall Takeaways and Conclusion

The actual results are a mixed bag of things that were surprisingly hard versus surprisingly easy. The easy was largely in ways that came down to how Meta was able to define the problem space. Communications generic and simple and quick enough to easily imitate and even surpass, no reputational or decision theoretic considerations, you can respond to existing metagame without it responding to you. Good times. The hard was in the tactical and strategic engines being lousy (relative to what I would have expected), which is more about Meta not caring or being skilled enough to make a better one rather than it being impossible.

Gwern notes that in June 2020 that Diplomacy AIs were a case of ‘the best NNs can’t even beat humans at a simplified Diplomacy shorn of all communication and negotiation and manipulation and deception aspects.’ I think this is selling the deceptive aspects of no-press (e.g. no communication) Diplomacy short, although it highlights that NNs have a terrible time anticipating human reactions in multiplayer settings, as well. Mostly it seems to me like a case of the people involved not trying all that hard, and in particular not being willing to do a bunch of kludges.

This blog post from Gary Marcus and Ernest Davis gives the perspective that this shows that Ai is not primarily about scaling, offering additional details on how Cicero works. There were a lot of distinct moving pieces that were deliberate human designs. This contrasts with Gwern’s claim that the scaling hypothesis predicted Diplomacy would fall whereas researchers working on the problem didn’t.

I think I come down more on Marcus’ side here in terms of how to update in response to the information. How it was done, in context, seems more important than who claimed it would get done how fast.

I do not get any points for predicting this would happen, since I did not think about the question in advance or make any predictions. It is impossible to go back and confidently say ‘I would have made the right prediction here’ after already knowing the answer. My guess is that if you’d asked, in the abstract, about Diplomacy in general, I would have said it was going to be hard, however if you’d told me the details of how these games were played I would have been much less skeptical.

I do know that I was somewhat confused how hard no-press Diplomacy was proving to be in previous attempts, or at least took it more as evidence no one was trying all that hard relative to how hard they tried at other problems.

I also note that there wasn’t much discussion that I saw of 2-player Diplomacy variations, of which there are several interesting ones, as a way of distinguishing between simultaneous play being difficult versus other aspects. Are Diplomacy actually surprisingly difficult? This would tell us. Perhaps I simply missed it.

Gwern’s conclusion in the comments of this post is that the main update from the Diplomacy AI is that Meta bothered to make a Diplomacy AI. This seems right to me, with the note that it should update us towards Meta being even more of a bad actor than we previously assumed. Also the note that previously Diplomacy had seemed to be proving surprisingly hard in some aspects, and that seems to have largely gone away now, so the update is indeed in the ‘somewhat scarier’ direction on net. Gwern then offers background and timeline considerations from the scaling hypothesis perspective.

My big picture takeaway is that I notice I did not on net update much on this news, in any direction, as nothing was too shocking and the surprises often cancelled out.



Read the whole story
denubis
8 hours ago
reply
Share this story
Delete

House prices tipped to fall further, rents to climb higher as rate rises bite

1 Share
ANZ economists believe house prices will fall further next year before a modest 2024 recovery even as the rental market tightens sharply.

Read the whole story
denubis
9 hours ago
reply
Share this story
Delete

Nvidia AI plays Minecraft, wins machine learning conference award

1 Share
MineDojo's AI can perform complex tasks in Minecraft.

Enlarge / MineDojo's AI can perform complex tasks in Minecraft. (credit: Nvidia)

A paper describing MineDojo, Nvidia's generalist AI agent that can perform actions from written prompts in Minecraft, won an Outstanding Datasets and Benchmarks Paper Award at the 2022 NeurIPS (Neural Information Processing Systems) conference, Nvidia revealed on Monday.

To train the MineDojo framework to play Minecraft, researchers fed it 730,000 Minecraft YouTube videos (with more than 2.2 billion words transcribed), 7,000 scraped webpages from the Minecraft wiki, and 340,000 Reddit posts and 6.6 million Reddit comments describing Minecraft gameplay.

From this data, the researchers created a custom transformer model called MineCLIP that associates video clips with specific in-game Minecraft activities. As a result, someone can tell a MineDojo agent what to do in the game using high-level natural language, such as "find a desert pyramid" or "build a nether portal and enter it," and MineDojo will execute the series of steps necessary to make it happen in the game.

Examples of tasks that MineDojo can perform.

Examples of tasks that MineDojo can perform. (credit: Nvidia)

MineDojo aims to create a flexible agent that can generalize learned actions and apply them to different behaviors in the game. As Nvidia writes, "While researchers have long trained autonomous AI agents in video-game environments such as StarCraft, Dota, and Go, these agents are usually specialists in only a few tasks. So Nvidia researchers turned to Minecraft, the world’s most popular game, to develop a scalable training framework for a generalist agent—one that can successfully execute a wide variety of open-ended tasks."

(credit: Nvidia)

The award-winning paper, "MINEDOJO: Building Open-Ended Embodied Agents with Internet-Scale Knowledge," debuted in June. Its authors include Linxi Fan of Nvidia and Guanzhi Wang, Yunfan Jiang, Ajay Mandlekar, Yuncong Yang, Haoyi Zhu, Andrew Tang, De-An Huang, Yuke Zhu, and Anima Anandkumar of various academic institutions.

You can see examples of MineDojo in action on its official website, and the code for MineDojo and MineCLIP is available on GitHub.

Read Comments

Read the whole story
denubis
18 hours ago
reply
Share this story
Delete

Data breach crackdown amendments pass parliament

1 Share

The Albanese government’s crackdown on corporates that “fail to take adequate care of customer data” has intensified.

The post Data breach crackdown amendments pass parliament appeared first on The Mandarin.

Read the whole story
denubis
19 hours ago
reply
Share this story
Delete

Bulldozer flattens $600m skills package

1 Share
Scott Morrison (Image: ABC)

Scott Morrison reportedly hated NSW Treasurer Matt Kean so much he refused to sign off on a $600 million skills project for his home state, and a UN report says the Great Barrier Reef should be listed as a world heritage site in danger.

The post Bulldozer flattens $600m skills package appeared first on Crikey.

Read the whole story
denubis
20 hours ago
reply
Share this story
Delete

2022-11-27 over the horizon radar pt I

1 Share

One if the most interesting things about studying history is noting the technologies that did not shape the present. We tend to think of new inventions as permanent fixtures, but of course the past is littered with innovations that became obsolete and fell out of production. Most of these at least get the time to become well-understood, but there are cases where it's possible that even the short-term potential of new technologies was never reached because of the pace at which they were replaced.

And of course there are examples to be found in the Cold War.

Today we're going to talk about Over-the-Horizon Radar, or OTH; a key innovation of the Cold War that is still found in places today but mostly lacks relevance in the modern world. OTH's short life is a bit of a disappointment: the most basic successes in OTH were hard-won, and the state of the art advanced rapidly until hitting a standstill around the '90s.

But let's start with the basics.

Radar systems can be described as either monostatic or bistatic, terms which will be important when I write more about air defense radar. Of interest to us now is monostatic radar, which is generally what you think of when someone just says "radar." Monostatic radars emit RF radiation and then observe for a reflection, as opposed to bistatic radars which emit RF radiation from one site and then receive it at another site, observing for changes. Actually, we'll see that OTH radar sometimes had characteristics of both, but the most important thing is to understand the basic principle of monostatic radar, of emitting radiation and looking for what bounces back.

Radar can operate in a variety of parts of the RF spectrum, but for the most part is found in UHF and SHF - UHF (Ultra-High Frequency) and SHF (Super High Frequency) being the conventional terms for the spectrum from 300MHz-3GHz and 3GHz-30GHz. Why these powers of ten multiplied by three? Convention and history, as with most terminology. Short wavelengths are advantageous to radar, because RF radiation reflects better from objects that are a large portion or even better a multiple of the wavelength. A shorter wavelength thus means that you can detect smaller objects. There are other advantages of these high frequencies as well, such as allowing for smaller antennas (for much the same reason, the gain of an antenna is maximized at multiples of the wavelength, or at least at divisions by small powers of two).

UHF and SHF have a disadvantage for radar though, and that is range. As a rule of thumb, the higher the frequency (and the shorter the wavelength), the shorter the distance it will travel. There are various reasons for this, a big one is that shorter wavelengths more readily interact with materials in the path, losing energy as they do so. This has been a big topic of discussion in 5G telephony; since some 5G bands are in upper UHF and lower SHF where they will not pass through most building materials. The atmosphere actually poses the same problem, and as wavelengths get shorter the molecules in the atmosphere begin to absorb more energy. This problem gets very bad at around 60GHz and is one of the reasons that the RF spectrum must be considered finite (even more so than suggested by the fact that, well, eventually you get visible light).

There's another reason, though, and it's the more important one for our purposes. It's also the atmosphere, but in a very different way.

Most of the time that we talk about RF we are talking about line-of-sight operations. For high-band VHF and above [1], it's a good rule of thumb that RF behaves like light. If you can see from one antenna to the other you will have a solid path, but if you can't things get questionable. This is of course not entirely true, VHF and UHF can penetrate most building materials well and especially for VHF reflections tend to help you out. But it's the right general idea, and it's very much true for radar. In most cases the useful range of a monostatic radar is limited to the "radio horizon," which is a little further away than the visible horizon due to atmospheric refraction, but not that much further. This is one of the reasons we tend to put antennas on towers. Because of the low curvature of the earth's surface, a higher vantage point can push the horizon quite a bit further away.

For air-defense radar applications, though, the type I tend to talk about, the situation is a little different. Most air-defense radar antennas are quite low to the ground, and are elevated on towers only to minimize ground clutter (reflections off of terrain and structures near the antenna) and terrain shadow (due to hills for example). A common airport surveillance radar might be elevated only a few feet, since airfields tend to be flat and pretty clear of obstructions to begin with. There's a reason we don't bother to put them up on big towers: air-defense radars are pointed up. The aircraft they are trying to detect are quite high in the air, which gives a significant range advantage, sort of the opposite situation of putting the radar in the air to get better range on the ground. For the same reason, though, aircraft low to the ground are more likely to be outside of radar coverage. This is a tactical problem in wartime when pilots are trained to fly "nap of the earth" so that the reverse radar range, from their perspective, is very small. It's also a practical problem in air traffic control and airspace surveillance, as a Skyhawk at 2000' above ground level (a pretty typical altitude here in the mountain west where the ground is at 6k already) will pass through many blind spots in the Air Force-FAA Joint Surveillance System.

This is all a somewhat longwinded explanation of a difficult problem in the early Cold War. Before the era of ICBMs, Soviet nuclear weapons would arrive by airplane. Airplanes are, fortunately, fairly slow... especially bombers large enough for bulky nuclear munitions. The problem is that we would not be able to detect inbound aircraft until they were quite close to our coasts, allowing a much shorter warning (and interception) time than you would expect. There are a few ways to solve this problem, and we put great effort into pursuing the most obvious: placing the radar sets closer to the USSR. NORAD (North American Air Defense Command) is a joint US-Canadian venture largely because Canada is, conveniently for this purpose, in between the USSR and the US by the shortest route. A series of radar "lines" were constructed across Alaska, Canada, and into Greenland, culminating with the DEW (Distant Early Warning) Line in arctic norther Canada.

This approach was never quite complete, and there was always a possibility that Soviet bombers would take the long route, flying south over the Pacific or Atlantic to stay clear of the range of North American radar until they neared the coasts of the US. This is a particularly troubling possibility since even today the population of the US is quite concentrated on the coasts, and early in the Cold War it was even more the case that the East Coast was the United States for most purposes. Some creative solutions were imagined to this problem, including most notably the Texas Towers, radar stations built on concrete platforms far into the ocean. The Texas Towers never really worked well; the program was canceled before all five were built and then one of them collapsed, killing all 28 crew. There was an even bigger problem with this model, though: the threat landscape had changed.

During the 1960s, bombers became far less of a concern as both the US and the USSR fielded intercontinental ballistic missiles (ICBMs). ICBMs are basically rockets that launch into space, orbit around to the other side of the planet, and then plunge back towards it at terminal velocity. ICBMs are fast: a famous mural painted on a blast door by crew of a Minuteman ICBM silo, now Minuteman Missile National Historic Park, parodies the Domino's Pizza logo with the slogan "Delivered worldwide in 30 minutes or less, or your next one is free." This timeline is only a little optimistic, ICBM travel time between Russia and the US really is about a half hour.

Moreover, ICBMs are hard to detect. At launch time they are very large, but like rockets (they are, after all, rockets, and several space launch systems still in use today are directly derived from ICBMs) they shed stages as they reach the apex of their trip. By the time an ICBM begins its descent to target it is only a re-entry vehicle or RV, and some RVs are only about the size of a person. To achieve both a high probability of detection and a warning time of better than a few minutes, ICBMs needed to be detected during their ascent. This is tricky: Soviet ICBMs had a tendency of launching from the USSR, which was a long ways away.

From the middle of the US to the middle of Russia is around 9000km, great circle distance. That's orders of magnitude larger than the range of the best extant radar technology. And there are few ways to cheat on range: the USSR was physically vast, with the nearest allied territory still being far from ICBM fields. In order to detect the launch of ICBMs, we would need a radar that could not only see past the horizon, but see far past the horizon.

Let's go back, now, to what I was saying about radio bands and the atmosphere. Below VHF is HF, High Frequency, which by irony of history is now rather low frequency relative to most applications. HF has an intriguing property: some layers of the atmosphere, some of the time, will actually reflect HF radiation. In fact, complex propagation patterns can form based on multiple reflections and refraction phenomenon that allow lucky HF signals to make it clear around the planet. Ionospheric propagation of HF has been well known for just about as long as the art of radio has, and was (and still is) regularly used by ships at sea to reach each other and coast stations. HF is cantankerous, though. This is not exactly a technical term but I think it gets the idea across. Which HF frequencies will propagate in which ways depends on multiple weather and astronomical factors. More than the complexity of early radio equipment (although this was a factor), the tricky nature of HF operation is the reason that ships carried a radio officer. Establishing long-distance connections by HF required experimentation, skill, and no small amount of luck.

Luck is hard to automate, and in general there weren't really any automated HF communications systems until the computer age. The long range of HF made it very appealing for radar, but the complexity of HF made it very challenging. An HF radar could, conceptually, transmit pulses via ionospheric propagation well past the horizon and then receive the reflections by the same path. The problem is how to actually interpret the reflections.

First, you must consider the view angle. HF radar energy reflects off of the high ionosphere back towards the earth, and so arrives at its target from above, at a glancing angle. This means of course that reflections are very weak, but more problematically it means that the biggest reflection is from the ground... and the targets, not far above the ground, are difficult to discriminate from the earth behind them. Radar usually solves this problem based on time-of-flight. Airplanes or recently launched ICBMs, thousands of feet or more in the air, will be a little bit closer to the ionosphere and thus to the radar site than the ground, and so the reflections will arrive a bit earlier. Here's the complication: in ionospheric propagation, "multipath" is almost guaranteed. RF energy leaves the radar site at a range of angles (constrained by the directional gain of the antenna), hits a large swath of the ionosphere, and reflects off of that swath at variable angles. The whole thing is sort of a smearing effect... every point on earth is reached by a number of different paths through the atmosphere at once, all with somewhat different lengths. The result is that time-of-flight discrimination is difficult or even impossible.

There are other complexities. Achieving long ranges by ionospheric propagation requires emitting RF energy at a very shallow angle with respect to the horizon, a few degrees. To be efficient (the high path loss and faint reflections mean that OTH radar requires enormous power levels), the antenna must exhibit a very high gain and be very directional. Directional antennas are typically built by placing radiating and reflecting elements some distance to either side of the primary axis, but for an antenna pointed just a few degrees above the horizon, one side of the primary axis is very quickly in the ground. HF OTH radar antennas thus must be formidably large, typically using a ground-plane design with some combination of a tall, large radiating system and a long groundplane extending in the target direction. When I say "large" here I mean on the scale of kilometers. Just the design and construction of the antennas was a major challenge in the development of OTH radar.

Let's switch to more of a chronological perspective, and examine the development of OTH. First, I must make the obligatory disclaimer on any cold war technology history: the Soviet Union built and operated multiple OTH radars, and likely arrived at a working design earlier than the US. Unfortunately, few resources on this history escaped Soviet secrecy, and even fewer have been translated to English. I know very little about the history of OTH radar in the USSR, although I will, of course, discuss the most famous example.

In the US, OTH radar was pioneered at the Naval Research Laboratory. Two early prototypes were built in the northeastern United States: MUSIC, and MADRE. Historical details on MUSIC are somewhat scarce, but it seems to have been of a very similar design to MADRE but not intended for permanent operation. MADRE was built in 1961, located at an existing NRL research site on Chesapeake Bay near Washington. Facing east towards the Atlantic, it transmitted pulses on variable frequencies at up to 100kW of power. MADRE's large antenna is still conspicuous today, about 300 feet wide and perhaps 100 feet tall---but this would be quite small compared to later systems.

What is most interesting about MADRE is not so much the radio gear as the signal processing required to overcome the challenges I've discussed. MADRE, like most military programs, is a tortured acronym. It stands for Magnetic-Drum Radar Equipment, and that name reveals the most interesting aspect. MADRE, like OTH radars to come, relied on computer processing to extract target returns.

In the early '60s, radar systems were almost entirely analog, particularly in the discrimination process. Common radar systems cleared clutter from the display (to show only moving targets) using methods like mercury acoustic delay lines, a basic form of electronic storage that sent a signal as a mechanical pulse through a tube of mercury. By controlling the length of the tube, the signal could be delayed for whatever period was useful---say one rotational period of the radar antenna. For OTH radar, though, data needed to be stored on multiple dimensions and then processed in a time-compressed form.

Let's explain that a bit Further. When I mentioned that it was difficult to separate target returns from the reflection of the earth, if you have much interest in radar you may have immediately thought of Doppler methods. Indeed, ionospheric OTH radars are necessarily Doppler radars, measuring not just the reflected signal but the frequency shift it has undergone. Due to multipath effects, though, the simple use of Doppler shifts is insufficient. Atmospheric effects produce returns at a variety of shifts. To discriminate targets, it's necessary to compare target positions between pulses... and thus to store a history of recent pulses with the ability to consider more than one pulse at a time. Perhaps this could be implemented using a large number of delay lines, but this was impractical, and fortunately in 1961 the magnetic drum computer was coming into use.

The magnetic drum computer is a slightly odd part of computer history, a computer fundamentally architected around its storage medium (often not only logically, but also physically). The core of the computer is a drum, often a fairly large one, spinning at a high speed. A row of magnetic heads read and write data from its magnetically coercible surface. Like delay tubes, drum computers have a fundamental time basis in their design: the revolution speed of the drum, which dictates when the same drum position will arrive back at the heads. But, they are two-dimensional, with many compact multi-track heads used to simultaneously read and write many bits at each drum position.

Signals received by MADRE were recorded in terms of Doppler shifts onto a drum spinning at 180 revolutions per second. The radar similarly transmitted 180 pulses per second (PRF), so that each revolution of the drum matched a radar pulse. With each rotation of the drum, the computer switched to writing the new samples to a new track, allowing the drum to store a history of the recent pulses---20 seconds worth.

For each pulse, the computer wrote 23 analog samples. Each of these samples was "range gated," meaning time limited to a specific time range and thus distance range. Specifically, in MADRE, each sample corresponded to a 455 nmi distance from the radar. The 23 samples thus covered a total of 10,465 nmi in theory, about half of the way around the earth. The area around 0Hz Doppler shift was removed from the returned signal via analog filtering, since it always contained the strong earth reflection and it was important to preserve as much dynamic range as possible for the Doppler shifted component of the return.

As the drum rotated, the computer examined the history of pulses in each range gate to find consistent returns with a similar Doppler shift. To do this, though, it was first necessary to discriminate reflections of the original transmitted pulse from various random noise received by the radar. The signal processing algorithm used for this purpose is referred to as "matched filtering" or "matched Doppler filtering" and I don't really understand it very well, but I do understand a rather intriguing aspect of the MADRE design: the computer was not actually capable of performing the matched filtering at a high enough rate, and so an independent analog device was built to perform the filtering step. As an early step in processing returns, the computer actually played them back to the analog filtering processor at a greatly accelerated speed. This allowed the computer to complete the comparative analysis of multiple pulses in the time that one pulse was recorded.

MADRE worked: in its first version, it was able to track aircraft flying over the Atlantic ocean. Later, the computer system was replaced with one that used magnetic core memory. Core memory was random access and so could be read faster than the drum, but moreover GE was able to design core memory for the computer which stored analog samples with a greater dynamic range than the original drum. These enhancements allowed MADRE to successfully track much slower targets, including ships at sea.

The MUSIC and MADRE programs produced a working OTH radar capable of surveiling the North Atlantic, and their operation lead to several useful discoveries. Perhaps the most interesting is that the radar could readily detect the ionospheric distortions caused by nuclear detonations, and MADRE regularly detected atmospheric tests at the NNSS despite pointing the wrong direction. More importantly, it was discovered that ICBM launches caused similar but much smaller distortions of the ionosphere which could also be detected by HF radar. This improved the probability of HF radar detecting an ICBM launch further.

AND THAT'S PART ONE. I'm going to call this a multi-part piece instead of just saying I'll return to it later so that, well, I'll return to it later. Because here's the thing: on the tails of MADRE's success, the US launched a program to build a second OTH radar of similar design but bigger. This one would be aimed directly at the Soviet Union.

It didn't work.

But it didn't work in a weird way, that leaves some interesting questions to this day.

[1] VHF is 30-300MHz, which is actually a pretty huge range in terms of characteristics and propagation. For this reason, land-mobile radio technicians especially have a tendency to subdivide VHF into low and high band, and sometimes mid-band, according to mostly informal rules.

Read the whole story
denubis
1 day ago
reply
Share this story
Delete
Next Page of Stories