OT: The Truth About DARPA

[snip]

The downside to this approach is that your 'research' will now be encumbered by DARPA and other Federal regulations (think ITAR). Where you might have been able to sell your petrol engine worldwide, including a version designed to meet DARPA specific requirements, if you take their money that may no longer be the case.

Better to sell your product to the world. Then, when the Pentagon wonders why Russian, French, Chinese and al Quaida armored vehicles seem to be much more maneuverable and efficient than the steam-powered Abrams tank, the CIA will provide them with the intelligence on your products. No doubt, your technology will be stolen and handed to US defense contractors, who will produce a militarized version of your petrol engine (i.e. more $$$$). But at least, you'll have the rest of the world market to yourself.

Until the US contractor lobbies Congress, gets you shut down and takes over your marketing channels to your foreign customers. So you'd better move your R&D off shore before they figure out what you're up to.

--
Paul Hovnanian  paul@hovnanian.com
----------------------------------------------------------------------
Have gnu, will travel.

DARPA does get many good results, some of which are well known, and many of which aren't. I've dealt with DARPA a few times over the years, and I've been very impressed with their people.

The original poster, I gather, has some idea involving networking. DARPA doesn't do much network R&D any more; that's a mature technology and the private sector is funding the necessary work. If his idea is that great, he needs to get a patent and look for venture capital.

(Although, weirdly, if you have something that really works and want to sell it to the Government, it's better to have a trade secret rather than a patent. The reverse is true in the private sector. See 28 USC 1498.)

John Nagle

are,

.

...?

.

Here's the juice: you sound and write like a crank (read John Baez's crackpot index

You are talking about engineering, but the style is the same as those who 'prove' relativity is inconsistent, etc. etc.

DARPA program managers probably get contacted by such people once a week. They do not believe you because you are probably wrong. You have not even said what you have developed. It is quite possible to say this without endangering your intellectual and industrial property.

They would believe a contractor who made the same claims because those claims would probably be right, simply because a contractor who made false claims would no longer be a contractor.

Of course you have managed to stir up the exceitable gaggle of "university of life self-made men" that inhabit this list, but do not let that encourage you.

illywhacker;

This is proposition is flawed. Let's take the case of Some Universally Acclaimed High Esteem Institution (SUAHEI). SUHAEI has historically obtained a reputation for occasionally producing spectacular results - although the last such spectacular result was achieved a couple of decades ago.

Now, there is a general impression in the public that SUAHEI employees are better than 'the crowd'. Supposing this is true: When did any particular individual achieve this extraordinary competence? Before they got employed by SUAHEI? After?

Unless somebody can answer this particular question, there is no basis to your argument.

Take the case of smart antennas. Parametric beamformers like MUSIC and ESPRIT were discussed for years in the late '90s (and maybe longer). Major playes like Siemens and the telecomm operators were involved in sponosring and developing these ideas.

If just *one* of the people involved had actually studied the basics of MUSIC, like I suggested in this post

the obvious conclusion - that these methods are far too fragile to be used in comm systems - would have been drawn within hours. As it were, maybe a decade worth of research was more or less wasted on trivialities.

Of course, the argument agians this is that "Major players like Siemens doesn't make such blunders". Maybe not. But the *people* at Siemens do.

This would be true in a *commercial* world. Nevertheless, I once saw a system based on - once again - MUSIC, developed for a commercial application in a demanding environment. The idea was based on a statement by somebody influential, that 'MUSIC is best!' Of course, this guy had long left the company and the remaining staff had never challenged the statement. The company had spent years developing a working prototype for this device.

The problem was that when they deployed it (it was to be towed behind a ship at sea), the device did not even find the sound of ship that towed it. At this point, these people asked me for opinions, and I quickly found a couple of discrepancies between the assumptions behind the model and the actual situation. The most important was that MUSIC treats all sources as point sources, whereas the ship was so close it had to be treated as a distributed source.

It's a trivial mistake that would have been discovered during a 5-minute pro&con brainstorm. But this company was a major player in the business, so everybody got caught by your very argument: "Players *that* big just don't make such basic blunders."

Again, *people* do.

As for the world of R&D - corporate R&D included - does not work the same way the commercial world does.

In the commercial world, failure to produce a result is reason to terminate a contract, and indeed, possibly fatal to the contractor's business.

In the R&D world, however, failure to deliver a result is, as often as not, the deciding argument _in_favour_ of extending the contract: there is always some next detail to investigate.

You only need to *avoid* asking the decisive questions, and you have a chain of projects lined up for years to come.

Rune

You appear to have ignored the word 'probably' throughout my post.

illywhacker;

e

...

No. You made the mistake of thinking it made a difference to a fundamentally flawed argument.

Rune

ose

Let's see:

1) you do not address the main point I am making, which is that the OP sounds like a crank;

2) you make many assertions that are not backed up by any evidence;

3) you roll out some tired and ludicrous 'working man's' cliches about research, again without evidence;

4) you talk of an well-regarded 'institute' that has not produced anything useful for ten years;

5) you introduce an anecdote that might or might not be true but which in any case is only a single example not a systemic failure;

6) you seem to think that an argument that is false remains false when its conclusion becomes probabilistic (which is mathematically untrue);

7) like several other posters, you have a conspiracy theory of the functioning of the R&D community;

What next? Can you achieve cold fusion like the OP?

illywhacker;

those

That's a subjective opinion of yours.

Assertion: Parametric sum-of-sines methods (MUSIC, ESPRIT, etc) are not useful for real-life work.

Evidence 1: Check out the published literature on ESPRIT, and check what applications are discussed, the affiliations of authors and sponsoring institutions.

Evidence 2: The trivial demo I linked to.

Conclusion: Any antenna processing based on MUSIC or ESPRIT will catastrophically fail when the system load (number of users) exceed some threshold that depends on the physics of the antenna.

Which ones?

Which one?

For obvious reasons, I can't disclose details about who, where and when. Whether you believe the story, make sure you believe and understand the reason why this particular system failed.

There are storues about people surviving free falls from

30000ft. I still prefer not to do the jump at all, and if I have to, to use a parchute.

No conspiracies. Observations.

Where did the OP mention cold fusion?

Rune

Riiggghhhttt. I'm sure you have specific examples of programs in mind? With details?

Cheers

Phil Hobbs

Of course.

Rewind the clock to the mid '70s, the midst of the cold war. There were 'sound sources' in mid ocean that somebody had a great interst in locating, identifying and tracking. There were acoustic systems in place to achieve those goals, but these systems were hampered by certain limitations, so the 'watchmen' could not quite get all the information about the sources that they would like to have.

In the late '70s somebody suggested to exploit the interaction between the sound emanated from the sources.

Hinich: Maximum likelihood estimation of a radiating source in a waveguide Journ. Ac. Soc. Am. vol 66 nr 2, august 1979, 480.

Bucker: Use of calculated sound fields and matched-field detection to locate sound sources in shallow waters Journ. Ac. Soc. Am. vol 59 no 2, p 368, February 1976

The ideas (known as "Source Localization by Matched Field Processing", MFP) looked elegant, at first glance, but the computational tools needed to investigate the ideas were not available at the time.

Throughout the '80s these computational tools were developed, and around 1986-87 one started to do the first field tests of MFP to see if the tricks worked.

The strategy of MFP was to first measure the oceanographic parameters, store them, and then use this information to extract the desired parameters about the source, from the measured data. Like using knowledge about the system function to estimate the excitation, from processed data.

The method did not work as one had hoped for, and in 1989 a paper was published where numerical simulations were done to establish how sensitive the method was to oceanographic variability:

Hamson and Heitmeyer: Environmental and system effects on source localization in shallow water by the mathced-field processing of a vertical array Journ. Ac. Soc. Am. vol 86 no 5, p 1950, November 1989

The conclusions from this paper was essentially that the MFP method

1) Was thrown off by variation in just about any parameter of the ocean or measurement set-up that could possibly vary 2) Every single parameter had to be known to at least 3, often 4, significant digits.

So up to, and including, the publicasion of the Hamson & Heitmeyer, MFP was a 'science; in the usual sense of the word.

However, the cold war was still going strong, and the 'sources' in question were as big nuisances as ever.

So no one wanted to pull the plug on MFP.

In 1991, the "Great Idea" of MFP arrives in the paper

Collins and Kuperman: Focalization: Environmental focusing and source localization Journ. Ac. Soc. Am, vol 90 nr 3, September 1991, p 1410

Here, the authors suggest that one estimate *both* the oceanography (=3D system function) *and* the source parameters (=3D excitation) from one set of measured data.

In essence, the problem statement is:

Given x + y =3D 1, find x and y.

The contents of the paper is essentially a simulation study:

1) The authors formulate a geometric scenario 2) The authors simulate corresponding measured data 3) The authors 'switch hats', to use the simulated data to estimate soure parameters 4) The authors set up a simulator (the same as was used to generate the data?) to search for the source parameters 5) The authors keep an eye of the progress of the simulation 6) The authors terminate the simulation when the results match 'sufficiently well'

So basically, the authors show that if one use a simulator that at least is similar to the simulator that generated the data to replicate the 'reference' data, and this simulator 'happens' to guess at parameters close to the original parameters, the data are similar.

To spoon-feed: The decisive question no one have asked, is whether this method of finding *both* the excitation

*and* the sytem function from one observation, holds. To me it is pretty obvious that it doesn't.

Incredibly, this paper was not only published, but it also recieved wide-spread attention. The last time I checked (around 2002), it was cited over 80 times, according to the Science Citation Index.

I could go on, and mention that the paper

Soares, Siderius and Jesus: "Source localization in a time-varying ocean waveguide", Journ. Ac. Soc. Am., V 112 no 5, p 1879, 2002

reports a field experiment where one measured oceanographic variability in as real-time as practically possible, and that one of the conclusions in the paper was that MFP just did not work if one used oceanography data older than about one hour.

Nevertheless, the authors stated that "MFP is robust".

But I think this is enough, for now.

Rune

I have no idea whether MFP does or does not work, but I do know that you understand very little about estimation. Given prior models of x and y, i.e. a probability distribution, x and y can be estimated subject to the condition x + y =3D 1, and the uncertainties in these estimates calculated. Maximum likelihood assumes minimal prior knowledge and produces maximal uncertainty, and is usually a stupid choice in the presence of significant prior knowledge, as in the case of oceanography.

illywhacker;

That might be true, formally. The problem is that

1) One has no idea if there is a source of interst present at all. 2) One has no idea where it might be. 3) One has no information about the source characteritsics 4) One has only gross knowledge about the oceanography (data are known to within 2, maybe 3, significant digits; oceanography is continuously changing.) 5) There are almost certainly sounds form 'uninteresting' sources present when the measurements are made.

All in all, MFP is a waste of time.

Rune

RCA's superconducting memory program in the early '60s.

Jerry

--
Engineering is the art of making what you want from things you can get.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Unlike relativity, the things that I proposed can be demonstrated with a prototype. I have a demonstration with Air Force Research Laboratory this Monday, in fact, but under unfavorable terms.

Also, if the situation were reversed, me being the Program Manager with Ph.D. in computer science purporting to fund "order-of-magnitude improvements" in computer networking, and some random person made the claim I made, it would take me about 60 seconds to see if that person should advance past the crank gate. If the person passes, for example, by answering a few questions about fundamental concepts of computer networking, then another 5 minutes might be allowed, then 15 minutes, then a full hour.

One could start with basic questions, like "poles are here in 4th- order system...please characterize zero-input-response...", "what stochastic model pertains to inter-packet delay by system with queues as you describe them...what is expected load at root of federated hierarchy if request distribution is as such...etc...what is your access control model...I am in China, using wireless link from laptop...explain to me in detail all the security attacks that could happen on path between my laptop and NSA in US...what recent development in India eliminates need for 'very probably' clause in determination of primality using Miller-Rabin.." Eventually the point would come where the PM either thinks the claimant is a crank, or think the claimant has not yet said anything ridiculous.

I think this is reasonable. It is easy and low risk. It takes less time to rout a person's mind in this manner than it does to regurgitate things that can be easily read on DARPA's web site, like how to respond to their solicitations, how to use the SBIR search tool, how to visit the federal gov's BizOpps site, how to apply for a Dun & Bradstreet business ID, etc.

And since we are on this subject, in the history of my interaction with these PM's, only one of them got to the part that matters. She stated the usual "You do realize what you're saying is a bit hard to swallow...", I said, "yes...we can get pass that by a way of your choosing...", then we had back and forth where she challenged me to explain my approach to the mobility problem. After it was over, she said "If I were you, I would keep trying with the other PM, sometimes it takes a while..." The "other PM", on the other hand, as well as his boss, his colleagues...I might as well been talking to the admin assistant...it was as if they were trying to avoid, completely, getting to the part that matters.

Just two days ago, someone from Air Force was doing the..."we've been doing this for years, the likelihood that you have...yada..thing..." so I said, "Well then, since you're that deep..there is probably some old problem that you have had lying around for last 2-3 years, one with which you are intimately familiar, that you could tell me about, that if I answered, would at least convince you that we should continue. If I answer poorly, I will never bother you again. But if I answer favorably, then I would like a clear path for funding that does not involve being fed head-first to one of your prime contrators. What problem do you offer?"

He paused, refused to give me a problem, then asked me to set up a demo for him and his colleagues.

This sounds promising, but the problem with these demos are one-way. The entire meeting is spent where one person is repeatedly saying, "there is nothing remarkable about what you're doing, while the other is repeatedly saying "...back up a bit...tell me more about these 64- bits..." or "..what dynamic programming technique are you using to reduce load in your distributed Dikjstra" "...say again how you ensure topological optimality?..." Of course it will happen that something I say is not new at all, since everyone else is at least thinking about it, and they will say, "...that's not new, someone over at Y.A.D.A. has started research on that..." One guy, one who spent most of the hour trying to learn as much as he could, said, "Ok, this is interesting work, but let's face it, if I had the opportunity to write a clean-slate stack, where I could start over completely from scratch, of course I would be able to solve these problems too."

At the end, the answer is always the same: "You have nothing remarkable, but we would be interested if you wold simply take the names of these three existing contractors, whom we are currently funding, and give your stuff to them, and try to work out a deal where they give you some of the money we are giving them. Maybe in a few years, you too, can engage us, but right now, all of our resources are pre-allocated to them. Sorry, this is the way it works, unless you want to wait 2-3 years for the other process, which is not guaranteed, and which you are obviously reluctant to do."

I thought it was evident from what I wrote. It's a clean slate protocol stack as well a couple of miscellaneous th>> I asked NSF if I made a proposal for a clean-slate stack that would

That's not what I am seeing. Both the largest contractors on the military research I referenced were issued partial stop-work orders in

2005 because it became painfully evident to all watching that the end result was no where near in sight:

But a few months later, both were back in the game. A US Gov official recently remarked that they would like to get away from such contractors from time to time and go with small entities who actually have breakthrough results, but it's hard, because the primes know how to work the system.

On the contrary, I have watched their writings over the years, and I see no reason to disregard what they say.

-Le Chaud Lapin-

Almost certainly false. Depth is constrained at least.

None?! Could the source be a supernova under the sea?

That may be so, but MFP does not make use of prior knowledge: it is maximum likelihood, as you said.

illywhacker;

Bathymetry isn't. There could be an unknown mound or through at the sea floor. Or the geological characteristics of the sea floor might change. If that happens, you're screwed.

And of course, the simulators can only handle so many effects. If there are some effect present that the simulator can't handle, you're screwed.

Oceanography is dynamic on a scale far beyond what you would imagine. There are all kinds of wave phenomena travelling on the internal boundaries, like thermoclines. These are not well understood or studied, mostly because people can't see them.

I've seen ROV navigation data that indicated the pilot flew the thing as if he was drunk. On-line video from the sea floor prooved that the guy flew as he was supposed to. The nav system used acoustic links from the surface vessel to the ROV, and were almost certainly screwed up by acoustic propagation paths being screwed up by some sort of internal wave in the water column.

Well, sources *have* been known to have similar characteristics. In practice, the decision tree goes something like

Q1) Is there a source present? If 'yes': Q2) Are the characteristica consistent with A) One of many sources of interest? B) One of many sources not of interest? If 'A': Q3) What are the parameters of interest for the source?

The hard question is Q2. Without going into detail, there are only so many characteristica to look for, and only a small subset decide the outcome of Q2. Even if one can decide Q2, you are basically screwed if there are other sources present with similar characteristica as the source of interest.

Rune

...

When a discovery is new, people say, "It isn't true." When it becomes demonstrably true, they say, "It isn't useful." Later, when its utility is evident, they say, "So what? It's old." a paraphrase of William James

Jerry

--
Engineering is the art of making what you want from things you can get.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

All sorts of people say stupid things in scientific papers. So what? We were talking about DARPA programs.

Nice try.

Cheers

Phil hobbs

--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058
hobbs at electrooptical dot net
http://electrooptical.net

Lots of stupid things get done. Done some myself. Was that a DARPA program?

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058
hobbs at electrooptical dot net
http://electrooptical.net