Sonar and AI revisited

[I'm goin' down]

Ducimus, of TMO fame, directed me to to a link explaining sonar and AI. It is a great tutorial. Here is Ducimus' post in another threat re subject. Follow the link.

Ducimus: "I keep reposting this. I wrote it some years ago for SH3, but the same principles apply. http://www.ducimus.net/sh415/ai.htm"

[SteveUK]

Ducimus come up tops again thanks for sharing this is quite a useful find

[I'm goin' down]

I think the powers that be need a sticky for tips like this one. I have been following the wrong strategy for months because I assumed freezing my position (like an ostrich with his head in a hole) was the proper and safe technique. Wrongamundo!

[Rockin Robbins]

Going into my Sub Skipper's Bag of Tricks #1 post later today! that will automatically sticky it. This post was what got me fast tracked on evasion strategy too, now that you mention it!

[I'm goin' down]

Wow, I have come a long way since you taught me how to use auto targeting. But I was always lagging. I never did understand how to evade, and it has been my bane.

After reading the tutorial above, I thought NOW IS THE TIME TO BRING IT. I was confident as I knew I now had the key to victory and a fighting chance for survival. I got so excited I played a private mission in my boat's namesake, the USS Barb, planning to evade, hide and pop up to kill the enemy. I anxiously set off into Formosa Strights where the Barb was promptly hit by active sonar pings from a D class destroyer which liesurely proceeded to lay down depth charges at deep depths badly damaging my new boat, including destroying both periscopes and killing 1/3 of the crew. Undaunted, I headed to Saipan for repairs. Pulling into port I managed to run aground and sink her at the mouth of the harbor. The remaining crew must have been happy to be home as they ran off the boat faster than I ever saw them move when at battle stations. I am sure I will receive a promotion or medal for bravery, as I just received an invitation for a private meeting with a bunch of admirals. It was hand delivered to me by two MPs. I owe it all to the tutorial.

[Sniper297]

I'm goin' down liked my diagram of a WWII destroyer's detection areas, here's the link to that;

http://www.subsim.com/radioroom/show...9&postcount=27

I don't know all that much about AI other than it's supposed to stand for "Artificial Intelligence" and in some games it really means "Artificial Ignorance". But I can tell you what I know about sonar. I was an Aviation Anti Submarine Warfare Tech, about 30 years ago so everything I know is by now both obsolete and declassified. My primary expertise was the AQS-13 Alpha dip sonar on the Sikorsky SH3D Sea King helicopter, but I also made a study of sound in general.

Passive sonar is merely listening for noise. Sound travels about 1100 feet per second through the air at sea level pressure and 59 degrees farenheit. It travels through the same temperature sea water of normal salinity at about 4800 feet per second. Next time you go swimming at a lake where there is water skiing going on, look for a ski boat zipping along about half a mile or further away, depending on atmospheric conditions and other ambient noise, you might or might not hear the ski boat. Now duck your head underwater, that high pitched whine you hear is the propeller of the ski boat - that's passive sonar, with your ears acting as hydrophones. Water is a much better conductor of sound than air, sound waves travel faster and much further. If there are several ski boats and some slower fishing boats, try to pick out which is which and you'll get an idea of the difficulty for the sonar operator.

A modern hydrophone is a bit more sophisticated, with a ring of "ears" and electronic circuits which decide which "ears" are getting the sound loudest and converting that info into a general bearing. The AQS-13 was a lot more sophisticated than the WWII sonar sets, it had a PPI scope which displayed light flashes on the bearing. Range was a factor of volume, a quiet sound would display as a light spot at the outer edge, a louder sound closer to the middle, the operator had to make an educated guess if he was hearing something close and quiet or loud and distant. The sonar sets we have for the subs in the game are unidirectional (correct for the period) so we have to manually rotate the dial for the loudest sound level and read the direction, modern sonar sets do that automatically since they're omnidirectional.

Active sonar is merely an echo, shout at a canyon wall and hear your own voice echoed back a few seconds later. For that a transducer sends out a "ping!" into the water, then the hydrophone works the same way as it does for passive sonar, only the sound it hears is the echo of the one that the transducer sent out. The time difference between the ping and the pong is the range, as electronic circuits became capable of measuring milliseconds and nanoseconds that became more accurate.

WWII subs had three sonar heads - one on each side of the keel under the forward torpedo room, one on deck above the forward torpedo room. They placed the sound heads there to get them as far from the sub's own machinery noises as possible. The sonar head on deck was the primary one used when submerged, being above water when the sub was surfaced it didn't work on the surface. When operating near the bottom in shallow water the two bottom heads were rigged in (retracted) to prevent wrecking them if the sub bottomed and only the deck one was used. Reason I mention that is because someone questioned if sonar worked on the surface, yes it does - sort of.

Main problem with active sonar is the double edged sword, you can get a more accurate range and bearing to the target by pinging him, but anyone else with a hydrophone can also hear that ping which gives them an approximate range and bearing to YOU without revealing themselves. Other problem is the ocean is a cluttered place, in shallow water sound will bounce off the bottom, in heavy seas you get a plethora of echoes from the troughs of all the waves. So the question, in real life can you get a range and bearing on a ship using active sonar when you're on the surface? Maybe, it depends. The two bottom sonar heads will work, if the seas are calm and the target isn't too far away you can get a pretty good echo off a surface ship while you're on the surface. With a moderate swell you would get an echo off the Yamato since the bottom sticks down further than the wave troughs provided it's not too far away. but you would have to pick it out of a zillion false echoes from the wave troughs nearby. In rough seas you would definitely need to submerge to get better sonar conditions, otherwise the reverberation would make it impossible to detect anything except waves further than a few hundred yards.

For both active and passive sonar all this is complicated by the ocean itself - it's a noisy, crowded place, and there's no consistency to it. Sound waves travel at different speeds depending on the temperature and the salinity - the game doesn't model this, but the real oceans have layers and streams of water that have different concentrations of salt, and sound waves will bounce off off those. (A whale can also sound like a sub, but they have an international recognition signal, surfacing and blowing a spout!)

So what's all this mean, for detecting and tracking surface ships, be aware that ranges and bearings are approximate, that's why they canceled the prewar doctrine about sonar only attacks and said "use the periscope". When I get a sonar contact I turn in that direction and wait, check the bearing change and double the angle - if the bearing changes 10 degrees starboard after 15 minutes, I change course 20 degrees starboard and continue to close. 15 minutes later I might do a single ping to get a range and bearing, then another one 15 minutes later to get a second one, draw a line through the two marks and that's the APPROXIMATE course of the target. Do an end around or whatever it takes to get in front of him, then visual observations to get a more accurate course and position the sub 1000-2000 yards off the track then wait for him to reach the firing point.

Little spoiler here for those using realistic targeting, rather than screw around with speed, range, angle on the bow, and firing solutions I just set speed zero, angle on the bow zero, rotate the periscope straight ahead and move the mast height on the stadimeter so the range reads 1000 yards, send range and bearing to TDC. Now I got a zero gyro angle for the bow tubes which won't change, wait for the target to come. Slow speed, 5-8 knots, my angle off is 10 degrees, medium speed 10-12 knots 12 degrees, higher speeds 15 degrees. So if he's coming from the left running medium speed I set the crosshairs of the periscope to 348, when he moves into the crosshair I fire. That's the correct way to do it for Sugar boats anyway, early in the war they didn't have a TDC, and using the "angle off" method the exact range doesn't matter.

So you hit him and now the cans are after you, how to evade? Putting in a request for transfer to shore duty as fast as possible don't work, I tried that. First thing is go deep, switch to the chart and use the compass to draw a five mile circle around you. Pick any direction other than the original convoy course, you need to get to the outer ring of that circle. If there's a temperature layer, get under that - the guy whispers "passing thermal layer", as near as I can tell the BT detector is on the keel, so you need to get your conning tower under that layer, go 50-60 feet deeper. Shallow water or no layer down to your operational depth, here he comes, ping ping ping. The jig is up, if you're IN the active sonar arc you're already detected, so there's no point trying to be quiet NOW. So ruin his firing solution, go to flank speed and change depth - if you're deep go up 100 feet, if you're shallow go down 100 feet, don't turn until the charges go off because you're trying to move as fast as possible to get out from underneath them. Hit J to kick out a decoy, he's detected you HERE so after the attack HERE is the first place he'll look for you again, with a little luck he'll find the decoy and focus on that. After the charges explode turn 90 degrees and slow to 1/3rd (or 2 knots if you're using the knot meter). Continue to head for the outside of the circle until he detects you again, then repeat the previous, go to full tilt boogie, hit J and change depth just before he drops, turn 90 degrees and slow down after the charges explode. Lather rinse repeat until you're outside of that five mile circle, if there have been no nearby depth charge attacks in the last hour you can come up to periscope depth and take a peek.

[sergei]

Good reading Sniper.

Originally Posted by Sniper297

When I get a sonar contact I turn in that direction and wait, check the bearing change and double the angle - if the bearing changes 10 degrees starboard after 15 minutes, I change course 20 degrees starboard and continue to close.

This is a very nice quick rule of thumb for closing on a sonar contact. I'll have to try it out next time I have the opportunity.

Originally Posted by Sniper297

Little spoiler here for those using realistic targeting, rather than screw around with speed, range, angle on the bow, and firing solutions I just set speed zero, angle on the bow zero, rotate the periscope straight ahead and move the mast height on the stadimeter so the range reads 1000 yards, send range and bearing to TDC. Now I got a zero gyro angle for the bow tubes which won't change, wait for the target to come. Slow speed, 5-8 knots, my angle off is 10 degrees, medium speed 10-12 knots 12 degrees, higher speeds 15 degrees. So if he's coming from the left running medium speed I set the crosshairs of the periscope to 348, when he moves into the crosshair I fire. That's the correct way to do it for Sugar boats anyway, early in the war they didn't have a TDC, and using the "angle off" method the exact range doesn't matter.

Again very useful to know. I like my S Boats, so I'll try it out. But I gotta ask. Does this assume a Mark 14 on fast setting?

Originally Posted by Sniper297

If there's a temperature layer, get under that - the guy whispers "passing thermal layer", as near as I can tell the BT detector is on the keel, so you need to get your conning tower under that layer, go 50-60 feet deeper.

Interesting. I dunno if the game models this, would be good to know.

Originally Posted by Sniper297

... here he comes, ping ping ping. The jig is up, if you're IN the active sonar arc you're already detected, so there's no point trying to be quiet NOW. So ruin his firing solution, go to flank speed and change depth

If they are searching on long scale (ping pause ping pause ping), they they may just be fishing. Stay cool.
As soon as they change to short scale (ping ping ping) that's when I assume they are gearing up for an attack run, and try to get the hell away from their solution.
EDIT - This last bit is not directed at you. You obviously know far more about sonar than me. May be helpful for new guys though!

Interesting post Sniper

[Sniper297]

Good question. I dabble in Sugar boats and others occasionally, mostly I play Sargo because it's the best middle of the road compromise - only two less torpedoes and two less tubes than the Gato but capable of operating in shallower harbors. Best way to find out is to play with automatic targeting and look at the dials, check the bearing compared to the gyro angle. A ship crossing your bow at a 90 degree track with the chosen torpedo type and speed setting, note the speed and then note the actual bearing when the gyro angle pointer hits zero, that's your angle off for that torpedo. Experiment with different ranges and you'll find the angle is the same at 500 yards or 3000 yards. I'm currently tinkering with getting visual pictures of what various ships look like at different ranges using just the vertical marks, "fills X number of divisions in high power" to get a quick general idea of the approximate range. That's one of them practice practice practice things, if you play for a while with the unrealistic settings and pay attention to the sight pictures you get at different ranges while using those, when you go to ultra realism you'll be better able to line up the horizontal crosshair with the waterline, glance at which hash mark the top of the mast hits, and quickly guesstimate the range within a few hundred yards. That's the theory anyway, I'm still working on it.

As for the thermal layer, the game does indeed model that, when I get a task force I get directly in front and dive down below the layer, wait until the lead escort passes overhead then come back up to periscope depth to shoot the Yamato. Obviously that works best if the lead escort is a few thousand yards ahead of the battlewagons and the layer isn't too deep, and the timing is rather tricky since you need to come up soon enough to get set to fire as the target passes by, but not too soon so the can don't detect you. Whenever there's time I always dive down before the attack to check the actual depth of the thermal layer and how long it takes to get back up to periscope depth. If possible I also trail convoys or task forces for better conditions, in other words I get ahead and slow down to "trail" from ahead until daylight or until the convoy moves into deeper water.

[sergei]

It's not the modeling of the layer, and subsequent sonar reduction I'm wondering about.

It was the next bit: 'as near as I can tell the BT detector is on the keel, so you need to get your conning tower under that layer, go 50-60 feet deeper.'

I always assumed that once my crewman said 'passing thermal layer' I was under it. Didn't realise I may have to go deeper to compensate for the placement of the bathythermograph.
And yes - I did have to look that spelling up

[I'm goin' down]

starting with the 5th post in this thread, there is a lot of great new information. Thanks for the input.

[Sniper297]

Strictly a guess based on a few experiences, "passing thermal layer" at 170 feet and getting ping ping ping wham wham wham at 200, then dropping down to 220 and no more ping pong paddling. I'm more of a hacker than a programmer, but what little I do know about programming I'm guessing that it's an on-off setting similar to the visual sighting AI subroutine - if you're at a depth of 40 feet the top of the conning tower can be seen, at 50 feet the tops of the periscope shears can be seen. I SUSPECT (don't know for sure) that it's designed like a sort of virtual surface layer where if part of the sub sticks up above the layer it can be "seen" by the sonar. That would be the simplest way to program something like that, if they were to make it completely realistic it would require a lot more ones and zeros which would slow gameplay to a crawl. Anyone feels ambitious you can test that by stirring up a bunch of tin cans, then find a layer and experiment with different depths to see where the actual cutoff line is compared to the reported depth of the thermal layer.

In real life a thermal layer has to have a fairly big and sudden temperature difference to actually reflect the sound waves, what you usually get is "refraction", similar to what you see if you stick a pencil into a bowl of water and the pencil appears to bend. The refraction bends the sound waves so you still get a target, it's just not where the sonar says it is. Trying to program THAT in a computer game would be nuts, you'd need 10 gigs of code just for all the possible variables in that alone!

[Arclight]

Originally Posted by sergei

Again very useful to know. I like my S Boats, so I'll try it out. But I gotta ask. Does this assume a Mark 14 on fast setting?

Sorry for being a smartass, but:

S-boats only carry Mark 10. Mark 14 wouldn't fit the small tubes.


On sonar:

You can reduce the chances of detection by presenting a smaller profile. I'm sure Sniper can give a better explanation, but the amount of energy (sound) reflected from the ping is a lot lower if it comes at the boat from behind or from the front. I guess an operator might pass it off as a false return if it's too weak.

I think the sonar on WWII destroyers could be rotated (maybe not), but not angled down. So the "detection cone" would project outward and down at a limited angle. In other words, a warship would not be able to detect something directly beneath it. The deeper you are, the sooner your persuer looses sonar contact.

Not sure how the layer is modeled, but I think it's a percentile chance. Say getting below it knocks 10 or 20% off. Considering the game keeps track of how much profile you present to the sound source, it may matter if parts of the boat are still above the layer.

[sergei]

Originally Posted by Sniper297

ping ping ping wham wham wham

That sounds familiar. Story of all my careers so far.
My trouble is that in theory I am a master of evasion. Know the theory, all the angles, this should be a cakewalk.
But in practice I comprehensively suck at it. My knowledge doesn't seem to translate into survival. Something always seems to catch me out.

Originally Posted by Arclight

S-boats only carry Mark 10. Mark 14 wouldn't fit the small tubes.

Yeah you got me on that one

Originally Posted by Sniper297

I SUSPECT (don't know for sure) that it's designed like a sort of virtual surface layer where if part of the sub sticks up above the layer it can be "seen" by the sonar.

That makes a lot of sense. I'll adjust my behavior accordingly.

[Bubblehead1980]

I have posted this before but will share again, because so far I have found it to be the best evasion move in Silent Hunter, esp in TMO, which is well known for it's smart and rabib but NOT unbeatable escorts, as some would tell you.

This method is the "Dykers Method" after Lt Cdr Tommy Dykers, Skipper of USS Jack on her first and highly successful patrols.I got wind of the Tactic from Admiral James Calvert's book "Silent Running" Calvert was the TDC operator on the Jack at the time and rose to be her XO by wars end.

Example:

You attack a convoy in a Gato class, sink a ship or two, then here comes the escorts, you order her deep, silent running, 1 knot(50 RPM and below helps, use the knot o meter to set speed to 1 knot) in a Gato test depth is 300, can go deeper but i tend to stop at 300 unless need to go deeper.

Ship continues in, soon begans to ping, ping, can tell its coming in and has you, wait, when ship is roaring overhead like a freight train, ready to the chronometer, go ahead flank and start time, other DD's and the one overhead will NOT hear you because the roar of the escort drowns your sound out.You hear splashes, your speed has most likely carried you out from under the pattern but keep at flank, wait for explosions.Judge, how close are they, which can be difficult to do for those of us who play without external camera, but you can get a feel for how close they are via sound, vibration, if the subs shakes, its close.If close, order a 50 foot increase of depth.Order all stop and left or right full rudder(whichever seems best, away from escorts last direction gaine from sonar) let the momentum from the burst of speed carry you through the turn when speed drops , order 1 knot again and leave rudder hard over.

Another run, ping, ping, ping...as escort gets close, rudder amidships, as it closes and hear it roaring, again order ahead flank and start timer...after the run, order all stop and rudder in the OPPOSITE direction you ordered on last run, so if went to port last run, order starboard and again let the momentum carry you through then back to 1 knot.

The idea is that you never stay in same direction for long, always making a slow, constant turn, which confuses even TMO AI after a couple runs.Of course enemy sonar conditions affect your ability to evade but this method has always worked for me, rarely take hits and even more rare to lose my boat using this method.


The actual Dykers method in RL is just a constant slow turn but in the game, esp TMO, the speed bursts are something i incorporated out of necessity but it does work.Hope this helps.

[Sniper297]

Hee-hee, I think what he meant was did I come up with those settings using the mark 14 at high speed. Obviously if you're trying to shoot a flying duck with a bow and arrow you're going to need to lead it more unless it's a rocket powered arrow. IIRC the mark 10 and mark 18 run at the same speed as the low speed setting for the mark 14, and trying it I confirmed the obvious, slower torpedo speeds and/or faster target speeds need more lead angle.



This is one of the sub school exercises with 5 knot merchants, there isn't a LOT of difference at that low speed but the angle off does change when you change the speed setting of the fish. Like I said the thing to do is practice with auto targeting on, but instead of loafing around with your mind in neutral while the game does all the work, take notes on things like the appearance of targets at different ranges and angles, the actual periscope bearing when the gyro angle hits zero, and what speeds various ships normally run at.

Aspect, yeah, don't matter whether it's radar, sonar, or visual if you're pointing directly toward or away you're less detectable simply because you're "smaller". Place a pen on top of the refrigerator sideways, back off about five feet and look at it. Now rotate the pen so the point is facing you, look at it again from the same distance, changes from a target 7 inches wide to a target 3/8ths of an inch wide. That's primarily helpful for avoidance of detection though, if they already GOT you detected they're not gonna lose you if you turn toward or away, in fact it's better to be at right angles when trying to avoid depth charges. On the surface it's a different story, changing aspect means the difference between their gunners trying to hit a target 300 feet wide or 30 feet wide.