Accuracy Part 3.

Interesting stuff Julian.
I never did work out how to obtain a truely random number.

 

no offence taken Graham/Paul

for some reason,I seem to view Oedipus and BD in the same light,as trolls,intent on causing arguements.The problem may well be mine,you are both quite right

 

Excellent post Graham. btw, it only costs $65 a year (about £35) to join the AES as an associate member with access to all the conference papers and journals online. I've downloaded a few papers and I think it's worth it.

Michael.

 

penance,
there are reams of papers by academics on the subject of generating random numbers. the trick in a game are twofold 1) getting a random number (well pseudo random) without overloading the processor and 2) making sure you can repeat the sequence of random numbers for action replays / scripted levels etc. both of these usually mean that you use some bastardised combination of a seed value mutated every time the random function is called by various logical and arithmetic operations. not truely random but it does the job.
cheers


julian

 

Julian
I only ever used seed with the time functions, like you say, pseudo random.
Probably the wrong thread for me to ask more

 

The problem for me is that I have seen nothing to convince me that I am wrong in saying that I have heard differences between things that shouldn't make a difference ie cables for instance. When I, or somebody else says this, the usual statement is 'Prove it with DBT'. I would love to take part in a DBT using Isolda vs DNM Reson for instance as I have only used those in the last 6 years and did change for a reason ie the Isolda sounded better to my ears.

So lets do a DBT, I know SideShowBob said he would manage one so why not?

 

I'd be up for it. We've mentioned doing this before though and I don't think we have the number of people or the time and equipment to run a DBT that would have any statistically valid results. We'd have to do it properly level matched to within 0.1dB and use some kind of ABX switch box (if we were doing speaker cables) and then run quite a few trials (depending on how many people were present) in order to get meaningful results.

It might be interesting. However, whatever the result it's fairly sure to just cause more acrimony from one or other or both sides of the objectivist/subjectivist camps. As such it's probably not a very productive thing to do.

The objectivist/subjectivist divide in hifi is just one of those things that exists and very rarely do people change sides. We just have to learn to tolerate each other a bit better

Michael.

 

the Isolda leaves the DNM for dead,night and day,chalk and cheese blah blah,the DNM is still very good though,but this 'roll off' reputation lives on imo

 

Great idea. This has been suggested a number of times, I remember offering to host it at least once, but we've never been able to get "enough" participants to attend.

 

I think you are right, but I still say let's organise one.

 

Whatever DBT is run, we're not going to be able to 'sort out' the big debate. However, does that matter? Surely, we only need enough trials to statistically prove to Lee that he can or can't hear the difference.

After all, we've all got different sensitivities of hearing and so we shouldn't care what others can or can't hear. It's our own ears that matter to each of us.

I'm not a statistician but wouldn't this make it much easier to conduct tests? Also another advantage of doing the test for just one person (in this case Lee) is that the person need only select the programme material that he listens to typically.

If you're using seconds or smaller, that's pretty damn random.

 

i think the basic problem is with the statement '2 cables can sound different'. subjectivists just want to leave it at that and take the statement as it is. objectivists want to add 'only if they don;t measure the same and we can prove it through blind abx / null / some other testing' on to the end.
the issue is more complex than this second addendum as there are a number of psycological factors which colour our perceptions when listening in the real world which aren't present in an abx test - perhaps the biggest of which is that you haven't paid money for the cable(s) used in a blind abx test.
now as a staunch subjectivist i wouldn;t debate any of the above however i would question the relavence of a blind abx to a real world listener and i treat the idea that choosing a component solely on the basis of test results or blind abx testng very dubiously indeed. i'd certainly be interested in the results of a blind abx test and would love to take part too however i'm unsure as to what use the results would be as they would tell me very little about the real world situation they are 'measuring'.
note: this is just my own personal opinion and not intended to provoke ire etc...


using timers in random number generators (the horizontal blank counter is a good one to use as it updates 500+ times per second for pal) is good unless you want repeatability as aforementioned.
cheers

julian

 

It's always worth checking Google uk.rec.audio group if you want to do a cable DBT, as regular contributor Stewart Pinkerton still offers his (unclaimed) £1,000 challenge and may well have all the needed stuff to perform the test. I'm sure he'd be glad to accept a challenger. Plus there is the £1,000 if you pass! And the opportunity to silence all the 'objectivists' and make them eat humble pie off their eggy faces etc. etc. Though DNM and Isolda may not meet the 0.1db requirements as they are very different (two extremes of speaker cables).

 

Julian,

If the only audible difference between cables is measurable frequency response stuff then we can save a lot of money. This is what makes a DBT worth while, if it is magic then magic costs.

BTW Win32 includes an API for accessing the CPUs clock counter, increasing at gigahertz rates, so the lower bits are probably quite random when sampled.

OTOH what is 'random'? The digits of Pi meet most criteria but could not be considered random, a sequence that went 1 1 1 1 1 1 is as likely as 3 2 1 5 3 4 but would raise eyebrows...

Paul

 

On the random number front (since I'm about "accuracy'd" out), computers can never generate truly random numbers. That's why the lottery numbers are chosen the way they are. Premium Bond numbers are (or were) apparently chosen according to the count of electrons hitting the anode in a valve (or something like that). If some computer system was used, no matter how apparently random, there's a risk people would find a pattern.

Michael.

 

Ivor Tiefenbrun owner (of about half of) Linn participates in a Double Blind Test.

Cross posted from pfm here.

 

paul,
i know what you are saying re measurements however just testing cables in isolation seems to me to be only half the picture. how a cable interacts with what it's plugged into should also be taken into account as well surely? still at the end of the day it's 'my' money and caveat emptor and all that.

sampling the low bits of the system clock is fine however when you are writing cross platform stuff it tends to get a little more complex as a ps2, x-box and game cube all run at different clock speeds and therefore the game would play quite differently on each one if that was used. also, again, repeatability is not possible. the hbl counter stuff was only really applicable in the days of the commodore 64, amiga, st etc. for the cross platform stuff i use either a marsenne twister generator to generate uniformly distributed random numbers or a simpler linear feedback shift register approach using trinomials. as michael says true random numbers are an impossibility - except for a few early pentiums but you can get pretty close as the period of the latter aproach is massive.
cheers

julian

 

And that's part of the problem. I chose Isolda because I felt it better. When I say this I'm told cables sound the same, but when I say I want to do a DBT I'm told I can't use my cables.

I'm really happy to do a DBT. In many ways I want to be proven wrong. But I do want to put my own stipulation on it. I feel those being tested should be the believers, because those that don't believe may just put down A for each test and therefore skew the results.

 

notaclue - the 0.1dB requirement doesn't apply to the cables, it applies to the test. If, without altering the volume, there are level differences between two cables then that's not a problem. All that needs to happen is that when you switch from A to B the volume is set correctly so that this level difference is cancelled out.

Now, don't read this as skewing the test. I'm sure that most people who believe cables make a difference believe that there's much more to it than small level differences so I think it's fair to remove these (if they exist) and concentrate on the other differences (if any). Afterall, if the only difference was a few fractions of a dB then you could just turn up the volume a little and save a lot of money!

Lee, it's quite difficult to deliberately skew the results without knowing whether it's A or B being played. That's the point of the blind test, it means that neither believers nor sceptics can skew the result of the test in their favour. In any case, in previous DBTs, it's been shown that there's no difference in the ability of sceptics and believers to tell cables apart.

Michael.

 

Possibly a missing word here - or you're telling us more about your personal proclivities that you realy wanted!

Remember the important rider - "if they have the same electrical [i.e. LRC] parameters", which ISTM is the problem with the Isolda/DNM comparison. So it may be worth trying to build a cheap cable to match the LRC of the Isolda. The alternative (and possibly less restrictive) criterion is to level match - across the full frequency range.

Another interesting test here, which may head more to address the "magic" issue, would be to try an ABX with DCT and non-DCT Isolda. IIRC, you have the non-DCT and I have the DCT - which you are welcome to borrow if you wish. And does Bottleneck still have his Goertz - Isolda without the matching networks? The other potential problem is that (at least in the case of interconnects) there has been the frequent assertion that it's actually the connectors that make the major difference. If so, the presence of an ABX box (which must have at least as big an effect as connectors) may be a major perturbation in the test environment - giving the believers the "ah but that was destroying the purity of the basic cable setup" getout!

(FWIW - I'm probably in the agnostics-shading-sceptical camp at present; I suspect that LRC is the principal factor in cables, but am quite ready to believe that there may be some dielectric effects and can see that some electromagnetic/mechanical interactions could make a difference too. The one thing I am sure about is that the effects of cables are grossly overrated! And harmonic distortions can be nice too in certain situations - although they are clearly inaccuracies).

Michael - just putting down AAAAAAAAA would be a prima facie case of skewing the result! It may be a bit more difficult to skew the result less obviously - I suppose memorising a ( pseudo- ) random sequence of As and Bs would be required prior to the test.