Re: [Flexradio] 16 versus 24 bit audio ... how do you make these measurements?
I use RMAA as your measuring engine. http://audio.rightmark.org/index_new.shtml We are trying for a good 16-17 bits total range from these 24 bit A/D's. 20*log10(2^16) = 96 dB. That said, we NEED the 24 bit A/D's to get those good 16 bits but we do not need to set the sound system to 24 bits to get the good 16-17 bits. Peter is trying to justify his programming preferences which include not using the more complex tools to open sound cards, for example using WAVEFORMATEXTENSIBLE in the MME interface is quite a bit more complex and if you do not need it and cannot justify it, why do it? On the other hand, those cards that use the highest end chips, such as the Lynx L22, Emu 1212M, and Audiophile 192, can all use larger than 16 bits and there is no intermediate step, you have to go all the way. Bob Everett Palmer wrote: Hello Peter, What software and/or hardware do you use to measure soundcards? Thanks, Everett, KG6RYB At 01:06 PM 5/19/2006, Peter Martinez wrote: From G3PLX: I am just starting to do some serious tests with my new 24-bit Firebox. With the audio input of a channel short-circuit, if I analyse the statistics of the 24-bit data, I find the bottom 8 bits are ALWAYS random. In fact the bottom 9-10 bits are random. This tells me there is no point at all in having 24-bit audio. I might just as well stay with 16-bits. Has anyone here done any rigorous tests comparing 16 and 24-bit audio and found any difference? Like patching the code to mask off the bottom 8 bits completely, or replacing the bottom 8 bits with random data. Maybe someone has added a checkbox to turn this sort of test feature on and off and see what, if any, is the difference in the speaker. 73 Peter -- AMSAT VP Engineering. Member: ARRL, AMSAT-DL, TAPR, Packrats, NJQRP/AMQRP, QRP ARCI, QCWA, FRC. ARRL SDR Wrk Grp Chairman Laziness is the number one inspiration for ingenuity. Guilty as charged! ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio ... how do you make these measurements?
Hello Peter, What software and/or hardware do you use to measure soundcards? Thanks, Everett, KG6RYB At 01:06 PM 5/19/2006, Peter Martinez wrote: From G3PLX: I am just starting to do some serious tests with my new 24-bit Firebox. With the audio input of a channel short-circuit, if I analyse the statistics of the 24-bit data, I find the bottom 8 bits are ALWAYS random. In fact the bottom 9-10 bits are random. This tells me there is no point at all in having 24-bit audio. I might just as well stay with 16-bits. Has anyone here done any rigorous tests comparing 16 and 24-bit audio and found any difference? Like patching the code to mask off the bottom 8 bits completely, or replacing the bottom 8 bits with random data. Maybe someone has added a checkbox to turn this sort of test feature on and off and see what, if any, is the difference in the speaker. 73 Peter ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
Hello Flexers, I am in route back from Dayton where was so good to see many of you our customers and to have many more join in the fun. I do want to add a couple of points that have been missed here. Virtually all professional audio cards use +4 dBu as the nominal input level. 0 dBu is 1 mW into 600 Ohms. As I recall, the peak level before overload on most cards will be 10-20 dB above that. Most of these pro audio sound cards actually use some attenuation on the front end to prevent overload and give headroom at the nominal level. Attenuation introduces noise. On the other hand, the attenuation can be used to our advantage because the SDR-1000 utilizes a very high dynamic range preamp ahead of the detector. The total system gain ahead of the sound card is about 40 dB. This allows us to have the high dynamic range while not overloading the sound card. Most 16 bit cards use -10 dBV nominal inputs, which will overload at much lower levels. The only number I pay much attention to on the sound card specs is the dynamic range as measured in 20 KHz. Note that you typically need to subtract 3dB from the A weighted spec to get a more accurate picture of how it will work in our application. The Delta 44 is rated at 99 dB A Weighted and the Firebox at 107 dB A Weighted. The maximum theoretical specification for a 16 bit card is 96 dB + 1.72 dB (going from memory so I could be wrong on the 1.72). Note that the Turtle Beach Santa Cruz was one of the best 16-bit consumer cards when it was available. One of the reasons for this was that they actually used an 18-bit converter but only presented 16-bits to the bus. Even with that, it did not quite reach 16-bits of real dynamic range. It also would overload at much lower levels. So the Firebox will be 10 dB better than a PERFECT 16 bit converter, which does not exist on the consumer market to my knowledge. Don't forget that 10 dB is a factor of 10. I will take a factor of 10 improvement any day. Heck 3dB is twice as good in my book. The AKM chip converter mentioned earlier is rated by the manufacturer with an A weighted dynamic range of 123 dB so drop 3 dB off of that an you are just a fraction over 20 bits (ENOB) with one of the best converters. 120 dB down from 5V is 5uV! Very good board and circuit design is required to achieve that dynamic range. This is about what the $1000 cards do (e.g. Lynx 22). By the way, they work great with the SDR-1000. ;) 73, Gerald Gerald Youngblood, K5SDR FlexRadio Systems Ph: 512-535-5266 Fax: 512-233-5143 Email: [EMAIL PROTECTED] Web: www.flex-radio.com ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
From G3PLX: Alberto: Thanks! That's the information I was seeking. It looks as if I was half-right, if there are, at best, 18 good bits in a 24-bit card. As Phil says, if 16-bit cards turn out to be worse that 16-bits, then there is some reason to go for a 24-bit card (which is worse than 24bits!). But for sure, 24 bit cards are not going to be 256-times (48dB, 8-bits) better than 16-bit cards. I will now continue working to convert my own 16-bit SDR code to 24-bits, but I think I will add a button on the front panel which zaps the bottom 8 bits. It will be VERY interesting to see if it makes any difference. To take Frank's point, as far as I can see from both the ASIO and WMME interfaces, there IS an accessible point in the software where this can be done before the audio data becomes floating-point. Thanks to all who replied. 73 Peter ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
Peter, One subtlety that was mentioned once but not emphasized very much is that some of what you may be seeing as noise is actually deliberate dithering in the card, which has the effect of pushing quantization noise out to inaudible frequencies. 8-bits seems a bit much for dithering to me, but still, a perfect non-dithering card, i.e., one that outputs a pure string of '0's for a shorted input, is arguably _worse_ than a dithering card that outputs a controlled spectrum of noise on shorted input. It can all get rather weird when there are non-linear, low S/S+N psychoacoustic detectors at the end of the chain, i.e., ears. Chris - AE6VK -Original Message- From: Peter Martinez [mailto:[EMAIL PROTECTED] Sent: Friday, May 19, 2006 11:20 PM To: Flex Reflector Subject: Re: [Flexradio] 16 versus 24 bit audio From G3PLX: Alberto: Thanks! That's the information I was seeking. It looks as if I was half-right, if there are, at best, 18 good bits in a 24-bit card. As Phil says, if 16-bit cards turn out to be worse that 16-bits, then there is some reason to go for a 24-bit card (which is worse than 24bits!). But for sure, 24 bit cards are not going to be 256-times (48dB, 8-bits) better than 16-bit cards. I will now continue working to convert my own 16-bit SDR code to 24-bits, but I think I will add a button on the front panel which zaps the bottom 8 bits. It will be VERY interesting to see if it makes any difference. To take Frank's point, as far as I can see from both the ASIO and WMME interfaces, there IS an accessible point in the software where this can be done before the audio data becomes floating-point. Thanks to all who replied. 73 Peter ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
From G3PLX: Chris: I wondered about dithering. If I understand the idea, they introduce some deliberate noise in the ANALOGUE part of the ADC in order to spread the spectrum of any quantisation noise. Without it, I can imagine a tiny sinewave input of, say, one bit amplitude, looking to the software like a one-bit squarewave, with 3rd, 5th, ... harmonics. Dithering the analogue input would add a little noise to the fundamental sinewave but push the harmonics down. But, as you say, surely they would only introduce about one bit's worth of dither noise, not eight? 6dB not 48dB? I just did some proper measurements on the Firebox. On runs of 1200 samples at 48kHz, the r.m.s. noise level on the line input is about 400. It's about 450 on the Mike input. I did this by squaring each sample, summing these, dividing by 1200, and taking the square root. At the same time I calculated the DC level, just to make sure that was a lot smaller and didn't confuse things. This is about 9 bits of noise, or more if you suppose the peaks are higher. On the built-in 16-bit card in this laptop the DC level is 15 and I am not sure how to allow for this in the r.m.s calculation, but only the lsb toggles randomly. The MP3+ is the same, unless I select the mike input when the rms noise level is about 10. These figures are what I would have expected intuitively. One would expect the analogue noise-level to be roughly the same as the quantisation noise, in the same way that one would optimise any receiver so that all sources of noise contribute more-or-less the same in order to maximise the dynamic range. Intuitively I would have expected a 24-bit card to do the same, but maybe I am being naive. These measurements, and the figure quoted by Alberto, tells me that the 24-bit figure in the spec. is to some extent a marketting ploy. I can imagine a scenario where the designer of a new soundcard measured that it only gave 18 bits of useful data, so he decides to mask the data to 18-bits and sell it as an 18-bit card. The marketting manager of the company would surely override him, especially if the competition were all actively marketting so-called 24bit cards! 73 Peter ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
Peter, I asked the same question in a different way a while back. I noticed that with no inputs to the Firebox (or Delta 44 - I have both) the analogue input meters on the SDR software were reading in the order of 100dB down on maximum. Doing some simple maths the theoretical minimum fro a 24bit sample is 145dB down so where is the other 45dB of noise coming from? I may be over simplifying things but something doesn't seem right? As for dithering I read an article recently (maybe from Analog devices - can't remember) and do remember them talking about 1bit dithering. What I was trying to ascertain is whether there is a hardware issue or if it's a software issue (conscious design decision or otherwise). Apologies that this hasn't helped answer anything, but at least one other on the group is asking the same question! Regards, Paul M1PAF -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Peter Martinez Sent: 20 May 2006 10:07 To: Flex Reflector Subject: Re: [Flexradio] 16 versus 24 bit audio From G3PLX: Chris: I wondered about dithering. If I understand the idea, they introduce some deliberate noise in the ANALOGUE part of the ADC in order to spread the spectrum of any quantisation noise. Without it, I can imagine a tiny sinewave input of, say, one bit amplitude, looking to the software like a one-bit squarewave, with 3rd, 5th, ... harmonics. Dithering the analogue input would add a little noise to the fundamental sinewave but push the harmonics down. But, as you say, surely they would only introduce about one bit's worth of dither noise, not eight? 6dB not 48dB? I just did some proper measurements on the Firebox. On runs of 1200 samples at 48kHz, the r.m.s. noise level on the line input is about 400. It's about 450 on the Mike input. I did this by squaring each sample, summing these, dividing by 1200, and taking the square root. At the same time I calculated the DC level, just to make sure that was a lot smaller and didn't confuse things. This is about 9 bits of noise, or more if you suppose the peaks are higher. On the built-in 16-bit card in this laptop the DC level is 15 and I am not sure how to allow for this in the r.m.s calculation, but only the lsb toggles randomly. The MP3+ is the same, unless I select the mike input when the rms noise level is about 10. These figures are what I would have expected intuitively. One would expect the analogue noise-level to be roughly the same as the quantisation noise, in the same way that one would optimise any receiver so that all sources of noise contribute more-or-less the same in order to maximise the dynamic range. Intuitively I would have expected a 24-bit card to do the same, but maybe I am being naive. These measurements, and the figure quoted by Alberto, tells me that the 24-bit figure in the spec. is to some extent a marketting ploy. I can imagine a scenario where the designer of a new soundcard measured that it only gave 18 bits of useful data, so he decides to mask the data to 18-bits and sell it as an 18-bit card. The marketting manager of the company would surely override him, especially if the competition were all actively marketting so-called 24bit cards! 73 Peter ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
From G3PLX: Thanks Paul. Thank goodness I am not alone! I just did a simple calculation. The ultimate rock-bottom noise power is kTB, where k is Boltzmann's constant (1.38e-23), T is room temperature (300 degrees K) and B is the bandwidth in which you measure it - let's say 24kHz for a 48kHz soundcard, so this works out to 1e-16 watts. Let's say the circuit is 600 ohms (just so we can get some idea of the picture), so the r.m.s. noise level across a 600 ohm source connected to a soundcard is going to be the square root of 600e-16, or 0.244uV. Again, just for the sake of it, let's say the soundcard ADC is designed for 2.2 volts peak-to-peak (that's 0dBm in 600 ohms). A 24-bit ADC would then give 2.2/(2^32) volts quantisation level. That comes out to 0.13uV. That's not far off the circuit noise level. That 'feels' about right to me. It says that an 'ideal' 24-bit soundcard should just be able to 'hear' it's own input noise. This is how we have always designed the front-ends of things! I am just throwing numbers in there, and the experts may say that the circuit isn't 600 ohm, or I haven't taken into account the noise factor of the pre-amp, or the typical ADC full-scale isn't 2.2 volts. But I am surely not wrong by a factor of 256? Where is all the noise coming from? Paul's observation that the noise is 100dB down on full-scale could just mean that there is that much noise coming from the SDR1000 RF hardware, but it would be easy to test this by unplugging it and short-circuiting the soundcard input. If the noise stays at the -100dB mark then it says the soundcard is that noisy, which is exactly what I find here with the Firebox. Maybe Paul has already done this. Note that the software measuring the noise needs to be set to the full soundcard bandwidth for this to make sense - it will look a lot cleaner if you select a narrow filter. I am not familiar with the SDR software so I don't know how this works. 73 Peter ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
Sorry Peter - should have said - these figures are with SDR hardware turned off and unplugged/inputs shorted out. -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Peter Martinez Sent: 20 May 2006 11:51 To: Flex Reflector Subject: Re: [Flexradio] 16 versus 24 bit audio From G3PLX: Thanks Paul. Thank goodness I am not alone! I just did a simple calculation. The ultimate rock-bottom noise power is kTB, where k is Boltzmann's constant (1.38e-23), T is room temperature (300 degrees K) and B is the bandwidth in which you measure it - let's say 24kHz for a 48kHz soundcard, so this works out to 1e-16 watts. Let's say the circuit is 600 ohms (just so we can get some idea of the picture), so the r.m.s. noise level across a 600 ohm source connected to a soundcard is going to be the square root of 600e-16, or 0.244uV. Again, just for the sake of it, let's say the soundcard ADC is designed for 2.2 volts peak-to-peak (that's 0dBm in 600 ohms). A 24-bit ADC would then give 2.2/(2^32) volts quantisation level. That comes out to 0.13uV. That's not far off the circuit noise level. That 'feels' about right to me. It says that an 'ideal' 24-bit soundcard should just be able to 'hear' it's own input noise. This is how we have always designed the front-ends of things! I am just throwing numbers in there, and the experts may say that the circuit isn't 600 ohm, or I haven't taken into account the noise factor of the pre-amp, or the typical ADC full-scale isn't 2.2 volts. But I am surely not wrong by a factor of 256? Where is all the noise coming from? Paul's observation that the noise is 100dB down on full-scale could just mean that there is that much noise coming from the SDR1000 RF hardware, but it would be easy to test this by unplugging it and short-circuiting the soundcard input. If the noise stays at the -100dB mark then it says the soundcard is that noisy, which is exactly what I find here with the Firebox. Maybe Paul has already done this. Note that the software measuring the noise needs to be set to the full soundcard bandwidth for this to make sense - it will look a lot cleaner if you select a narrow filter. I am not familiar with the SDR software so I don't know how this works. 73 Peter ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
Hi Peter, Let me assure you: 24 bits is better than 16. Exactly how much better obviously depends on lots of things. And 16 bits would indeed be plenty if there were 16-bit sound cards that could actually deliver that resolution. If your low 8 bits are truly random with no input, that's a good thing! It follows that when you inject a non-random signal, it will stand out clearly from white noise. If the ADC resolution is good enough, this will also work at extremely low input levels i.e. when the S/N ratio is negative. Everyone, please read the specs of your sound cards. I don't have the PreSonus, but I have been using the 24-bit Delta 44 for years, already before there was SDR-1000. The input dynamic range for Delta 44 is only 99 dB(A). Of course, dynamic range does not necessarily translate to effective number of bits. Nevertheless, if we calculate ENOB using this value, we get ENOB = (99 - 1.76) / 6.02 = 16.2 bits. Then, there's also the dB(A) issue. Using A-weighting for audio applications may (or may not) make some sense, but for SDR-1000 it generally does not. For example, it looks like my Delta 44 has been tweaked to give optimal noise performace between 1-10 kHz. Above 10 kHz it's several dB's worse. Which is a problem because the PowerSDR software uses a 11.25 kHz IF. 73, Sami OH2BFO ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
On 5/20/06, Sami Aintila [EMAIL PROTECTED] wrote: software uses a 11.25 kHz IF. Ehm... correction: that's 11.025 kHz. ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
On 5/20/06, Peter Martinez [EMAIL PROTECTED] wrote: from G3PLX: Hello Sami. Thanks for joining the discussion - I feared I was on my own for a while! I hope some others will contribute too, so I don't get accused of hogging the bandwidth (or perhaps generating too much noise!). You said: If your low 8 bits are truly random with no input, that's a good thing! It follows that when you inject a non-random signal, it will stand out clearly from white noise. Let me be simple-minded and respond:- But if my low 8 bits are truly random and I inject an 8-bit sinewave, isn't it then level with the noise? If I could reduce my 8 bits of added noise to, say, 4 bits, wouldn't my sinewave then be 24dB above the noise? I don't see how adding that much noise can ever be a good thing to do. I am not trying to trap you, I really would like to understand what's going on here. This is all new stuff to me. I had a private email from a broadcast engineer who confirmed what I was saying about dither noise. He said they added 0.5 bits-worth of white noise to the analogue signal before digitisation, and subjectively that was the best result. But that's half a bit of dither, not 8 bits. 73 Peter Really, its like we are continuing to beat a dead horse here... a 24 bit card is not going to give you the theoretical 24 bits of resolution, just like a 16 bit card does not necessarily give you its theoretical 16 bits of resolution. Now some 16 bit cards can come close, but most 24 bit cards under $1000 are really going to give you 18 ENOB. Go back and do a search of the Flex Forum... there are examples of so called 24 bit cards that are much worse than some of the better 16 bit cards. When they say 16 bits or 24 bits, the manufacturers are refering to the ADCs rated resolution stated by its manufacturer. The design of the ADC front end, board layout, power supply filtering, etc... determines whether the board will even come close to the ADC rated specs (which are sometimes very optimistic in the first place). The Delta 44 became the recommended card for the SDR-1000 for a long time because in its price range, its performance was good, it has two ins and two outs, and ASIO drivers were available for it. Phil Harmon has just completed testing of the AKC ADC in the JANUS ADC/DAC board. He is seeing great numbers... this is the ADC used in one of the $1000 cards. We are not going to see 24 bits, but it will be much better than the Delta 44 or the Firebox. 73 de Phil N8VB ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
On 5/20/06, Philip Covington [EMAIL PROTECTED] wrote: On 5/20/06, Peter Martinez [EMAIL PROTECTED] wrote: from G3PLX: Hello Sami. Thanks for joining the discussion - I feared I was on my own for a while! I hope some others will contribute too, so I don't get accused of hogging the bandwidth (or perhaps generating too much noise!). You said: If your low 8 bits are truly random with no input, that's a good thing! It follows that when you inject a non-random signal, it will stand out clearly from white noise. Let me be simple-minded and respond:- But if my low 8 bits are truly random and I inject an 8-bit sinewave, isn't it then level with the noise? If I could reduce my 8 bits of added noise to, say, 4 bits, wouldn't my sinewave then be 24dB above the noise? I don't see how adding that much noise can ever be a good thing to do. I am not trying to trap you, I really would like to understand what's going on here. This is all new stuff to me. I had a private email from a broadcast engineer who confirmed what I was saying about dither noise. He said they added 0.5 bits-worth of white noise to the analogue signal before digitisation, and subjectively that was the best result. But that's half a bit of dither, not 8 bits. 73 Peter Really, its like we are continuing to beat a dead horse here... a 24 bit card is not going to give you the theoretical 24 bits of resolution, just like a 16 bit card does not necessarily give you its theoretical 16 bits of resolution. Now some 16 bit cards can come close, but most 24 bit cards under $1000 are really going to give you 18 ENOB. Go back and do a search of the Flex Forum... there are examples of so called 24 bit cards that are much worse than some of the better 16 bit cards. When they say 16 bits or 24 bits, the manufacturers are refering to the ADCs rated resolution stated by its manufacturer. The design of the ADC front end, board layout, power supply filtering, etc... determines whether the board will even come close to the ADC rated specs (which are sometimes very optimistic in the first place). The Delta 44 became the recommended card for the SDR-1000 for a long time because in its price range, its performance was good, it has two ins and two outs, and ASIO drivers were available for it. Phil Harmon has just completed testing of the AKC ADC in the JANUS ADC/DAC board. He is seeing great numbers... this is the ADC used in one of the $1000 cards. We are not going to see 24 bits, but it will be much better than the Delta 44 or the Firebox. 73 de Phil N8VB P.S. I was so jaded by all this ADC bit stuff that when discussions were going on in the old Xylo group about which ADC to use, I lobbied hard for the TI PCM4202 since it was easy to get compared to the AKM AK5394A part and I was convinced that the performance would not be all that much better. Well, I was definitely wrong about that one... this came out in testing... A few very smart individuals were suggesting all along that we look at the AKM part. I am glad that Phil VK6APH and Bill KD5TFD continued to test all possibilities... 73 de Phil N8VB ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
Peter Martinez wrote: snip Where is all the noise coming from? Paul's observation that the noise is 100dB down on full-scale could just mean that there is that much noise coming from the SDR1000 RF hardware, but it would be easy to test this by unplugging it and short-circuiting the soundcard input. If the noise stays at the -100dB mark then it says the soundcard is that noisy, which is exactly what I find here with the Firebox. Maybe Paul has already done this. Note that the software measuring the noise needs to be set to the full soundcard bandwidth for this to make sense - it will look a lot cleaner if you select a narrow filter. - snip - Yes, this last sentence is very important. When you evaluate the noise level on a spectrum display, don't forget that you are evaluating it in a bandwidth equal to the FFT bin size. And you can see it easily for yourself. If the program allows for this, change the bin size, e.g. make it half, and the base level of noise will drop by 3 dB... 73 Alberto I2PHD ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
At 02:07 AM 5/20/2006, Peter Martinez wrote: From G3PLX: I just did some proper measurements on the Firebox. On runs of 1200 samples at 48kHz, the r.m.s. noise level on the line input is about 400. It's about 450 on the Mike input. I did this by squaring each sample, summing these, dividing by 1200, and taking the square root. At the same time I calculated the DC level, just to make sure that was a lot smaller and didn't confuse things. This is about 9 bits of noise, or more if you suppose the peaks are higher. On the built-in 16-bit card in this laptop the DC level is 15 and I am not sure how to allow for this in the r.m.s calculation, but only the lsb toggles randomly. The MP3+ is the same, unless I select the mike input when the rms noise level is about 10. These figures are what I would have expected intuitively. One would expect the analogue noise-level to be roughly the same as the quantisation noise, in the same way that one would optimise any receiver so that all sources of noise contribute more-or-less the same in order to maximise the dynamic range. Calculate the average of all the samples, then subtract that from the samples, then square, sum, and root. Or, if you have a statistics package, calculate the standard deviation of the input samples. That will be the rms noise voltage. Of more interest is actually running a clean sine wave in, and then looking at the rms noise power as a function of the digitized value. Analog input noise shows up as noise that is independent of where in the cycle it's sampled. Clock jitter shows up as increased noise near the zero crossings (where the derivative of the input is highest). James Lux, P.E. Spacecraft Radio Frequency Subsystems Group Flight Communications Systems Section Jet Propulsion Laboratory, Mail Stop 161-213 4800 Oak Grove Drive Pasadena CA 91109 tel: (818)354-2075 fax: (818)393-6875 ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
At 04:35 AM 5/20/2006, Peter Martinez wrote: from G3PLX: Hello Sami. Thanks for joining the discussion - I feared I was on my own for a while! I hope some others will contribute too, so I don't get accused of hogging the bandwidth (or perhaps generating too much noise!). You said: If your low 8 bits are truly random with no input, that's a good thing! It follows that when you inject a non-random signal, it will stand out clearly from white noise. Let me be simple-minded and respond:- But if my low 8 bits are truly random and I inject an 8-bit sinewave, isn't it then level with the noise? If I could reduce my 8 bits of added noise to, say, 4 bits, wouldn't my sinewave then be 24dB above the noise? I don't see how adding that much noise can ever be a good thing to do. I am not trying to trap you, I really would like to understand what's going on here. This is all new stuff to me. I had a private email from a broadcast engineer who confirmed what I was saying about dither noise. He said they added 0.5 bits-worth of white noise to the analogue signal before digitisation, and subjectively that was the best result. But that's half a bit of dither, not 8 bits. One might add a few more bits of noise if one wasn't very sure of one's A/D linearity performance, and if it didn't impair the overall performance. If you're oversampling, for instance, and going to reduce a 96 ksps down to, say, 24 ksps, the power from the added noise would be reduced by the averaging you're doing with the sample rate reduction. Considering that most people will be filtering the output of the A/D to, say, 20Hz-20Khz, this might not be a bad design tradeoff. ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
From G3PLX: Sami is right. You CAN recover a signal from noise by narrowing the bandwidth, so if my signal was level with the noise in 24kHz bandwidth, I could filter it to a narrower bandwidth (in software) and improve it's SNR. But that's true regardless of the number of bits in the raw data. So long as we have at least half an lsb of additive noise to dither away the quantisation problem, we can always gain SNR by reducing the downstream software bandwidth. Any more added noise at the front end than this only makes things - well - er - noisier. The noise power calculation I did earlier shows that there is ALREADY about the right amount of noise inherent in the physics to get the dither optimum for a 24-bit ADC at 0dBm. If there are cards with more noise than this, it's probably because the designer couldn't get it any lower for the price, not because he chose to add more noise for some subtle reason. To take Jim's points, I have only so far measured the rms noise with no input, and not yet looked to see if there are any clues in it's spectrum - these things would take a lot more time. Clock jitter wouldn't explain what I see (an output with no input) although it would certainly cause noise in the presence of a large pure tone. I haven't tried that yet either. Let me close this topic before Phil accuses me of cruelty to dead horses. Before I aquired a 24-bit card, I honestly believed that 24-bit cards would be 8 bits better than 16 bit cards. When I did get one recently, I was surprised to find this wasn't the case. Jim is right. 24 bit cards may only be slightly better than 16-bit cards. I have learned something this week. 73 Peter ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
On 5/20/06, Peter Martinez [EMAIL PROTECTED] wrote: From G3PLX: Let me close this topic before Phil accuses me of cruelty to dead horses. Before I aquired a 24-bit card, I honestly believed that 24-bit cards would be 8 bits better than 16 bit cards. When I did get one recently, I was surprised to find this wasn't the case. Jim is right. 24 bit cards may only be slightly better than 16-bit cards. I have learned something this week. 73 Peter Hi Peter, Well, I may have been a little too strong in making that statement (dead horse)...it was early morning here...no coffee yet consumed... etc... ;-) The FlexRadio Forum has some interesting discussions in the past about different sound cards and what to expect. It is pretty much true that some 24 bit cards are marginally better than some 16 bit cards. It also would be accurate to say that some 24 bit cards are worse than some of the better 16 bit cards. 73 de Phil N8VB ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
Thanks to everybody for the most interesting and educating discussion. Nevertheless, I am not happy until the JANUS version with AK5394A is in my hands and I have modified the QSD and the following amplifier. Best regards and special thanks to the HPSDR group for the good work done so far, Ahti OH2RZ On 20/05/06, Philip Covington [EMAIL PROTECTED] wrote: On 5/20/06, Peter Martinez [EMAIL PROTECTED] wrote: From G3PLX: Let me close this topic before Phil accuses me of cruelty to dead horses. Before I aquired a 24-bit card, I honestly believed that 24-bit cards would be 8 bits better than 16 bit cards. When I did get one recently, I was surprised to find this wasn't the case. Jim is right. 24 bit cards may only be slightly better than 16-bit cards. I have learned something this week. 73 Peter Hi Peter, Well, I may have been a little too strong in making that statement (dead horse)...it was early morning here...no coffee yet consumed... etc... ;-) The FlexRadio Forum has some interesting discussions in the past about different sound cards and what to expect. It is pretty much true that some 24 bit cards are marginally better than some 16 bit cards. It also would be accurate to say that some 24 bit cards are worse than some of the better 16 bit cards. 73 de Phil N8VB ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
Message: 39 Date: Sat, 20 May 2006 16:09:52 - From: Peter Martinez [EMAIL PROTECTED] Subject: Re: [Flexradio] 16 versus 24 bit audio To: Flex Reflector FlexRadio@flex-radio.biz Message-ID: [EMAIL PROTECTED] Content-Type: text/plain; format=flowed; charset=iso-8859-1; reply-type=response From G3PLX: Sami is right. You CAN recover a signal from noise by narrowing the bandwidth, so if my signal was level with the noise in 24kHz bandwidth, I could filter it to a narrower bandwidth (in software) and improve it's SNR. I've been wondering about this lately; Isn't this getting close to what the definition of process gain is? Incidentally, Analog Devices has some excellent white papers that go into the details of the tradeoffs between sensitivity, dynamic range, resolution, etc. Definitely recommended reading. But that's true regardless of the number of bits in the raw data. So long as we have at least half an lsb of additive noise to dither away the quantisation problem, we can always gain SNR by reducing the downstream software bandwidth. Any more added noise at the front end than this only makes things - well - er - noisier. The noise power calculation I did earlier shows that there is ALREADY about the right amount of noise inherent in the physics to get the dither optimum for a 24-bit ADC at 0dBm. If there are cards with more noise than this, it's probably because the designer couldn't get it any lower for the price, not because he chose to add more noise for some subtle reason. To take Jim's points, I have only so far measured the rms noise with no input, and not yet looked to see if there are any clues in it's spectrum - these things would take a lot more time. Clock jitter wouldn't explain what I see (an output with no input) although it would certainly cause noise in the presence of a large pure tone. I haven't tried that yet either. It would be interesting to see the spectral characteristics of the noise. Let me close this topic before Phil accuses me of cruelty to dead horses. Before I aquired a 24-bit card, I honestly believed that 24-bit cards would be 8 bits better than 16 bit cards. When I did get one recently, I was surprised to find this wasn't the case. Jim is right. 24 bit cards may only be slightly better than 16-bit cards. I have learned something this week. Seems to me that the marketing folks take the number of bits in the ADC and run with it. A 16 bit card is only 14 bits, and a 24 bit card is only 18 bits, so the 24 bit card is 4 bits better than the 16 bit card. Cheers, -- Larry Gadallah, VE6VQ/W7 lgadallah AT gmail DOT com -- next part -- An HTML attachment was scrubbed... URL: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/attachments/20060520/eadbfa34/attachment.html ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
[Flexradio] 16 versus 24 bit audio
From G3PLX: I am just starting to do some serious tests with my new 24-bit Firebox. With the audio input of a channel short-circuit, if I analyse the statistics of the 24-bit data, I find the bottom 8 bits are ALWAYS random. In fact the bottom 9-10 bits are random. This tells me there is no point at all in having 24-bit audio. I might just as well stay with 16-bits. Has anyone here done any rigorous tests comparing 16 and 24-bit audio and found any difference? Like patching the code to mask off the bottom 8 bits completely, or replacing the bottom 8 bits with random data. Maybe someone has added a checkbox to turn this sort of test feature on and off and see what, if any, is the difference in the speaker. 73 Peter ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
Peter, You might want to check out: http://hpsdr.org/wiki/index.php?title=HpsdrWiki:Community_Portal and hpsdr.org under the JANUS section. The JANUS board should give better performance than any sound card that you can buy under $1000. 73 de Phil N8VB On 5/19/06, Peter Martinez [EMAIL PROTECTED] wrote: From G3PLX: I am just starting to do some serious tests with my new 24-bit Firebox. With the audio input of a channel short-circuit, if I analyse the statistics of the 24-bit data, I find the bottom 8 bits are ALWAYS random. In fact the bottom 9-10 bits are random. This tells me there is no point at all in having 24-bit audio. I might just as well stay with 16-bits. Has anyone here done any rigorous tests comparing 16 and 24-bit audio and found any difference? Like patching the code to mask off the bottom 8 bits completely, or replacing the bottom 8 bits with random data. Maybe someone has added a checkbox to turn this sort of test feature on and off and see what, if any, is the difference in the speaker. 73 Peter ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
From G3PLX: Thanks for the pointer Phil. I guess hpsdr.org might be an interesting place to browse anyway, but maybe I didn't make my point clear. I am not complaining that the Firebox I have is NBG, I am wondering if anyone has done any tests which establish whether 24 bit audio gives a better radio than 16-bit audio. OK, I can accept that a good quality soundcard will probably give better results than a cheapo, and the good ones may well be 24-bit while the cheap ones are 16-bit, but has anyone ever masked off the bottom 8 bits of a good 24-bit card and detected the difference? 73 Peter ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
On 5/19/06, Peter Martinez [EMAIL PROTECTED] wrote: From G3PLX: Thanks for the pointer Phil. I guess hpsdr.org might be an interesting place to browse anyway, but maybe I didn't make my point clear. I am not complaining that the Firebox I have is NBG, I am wondering if anyone has done any tests which establish whether 24 bit audio gives a better radio than 16-bit audio. OK, I can accept that a good quality soundcard will probably give better results than a cheapo, and the good ones may well be 24-bit while the cheap ones are 16-bit, but has anyone ever masked off the bottom 8 bits of a good 24-bit card and detected the difference? 73 Peter Hi Peter, Have you looked at how many bits are noise free on a good 16 bit card? 73 de Phil N8VB ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
Alan K2WS demonstrated the horrendous broadband noise coming from a very expensive Audigy 2ZS card by using the spectrum display of his IC-7800 http://www.n9vv.com/K2WS.html He dumped the Audigy and bought a nice Delta-66 that has continued to delivery great performance and a clean TX signal. de ken n9vv Philip Covington wrote: On 5/19/06, Peter Martinez [EMAIL PROTECTED] wrote: From G3PLX: Thanks for the pointer Phil. I guess hpsdr.org might be an interesting place to browse anyway, but maybe I didn't make my point clear. I am not complaining that the Firebox I have is NBG, I am wondering if anyone has done any tests which establish whether 24 bit audio gives a better radio than 16-bit audio. OK, I can accept that a good quality soundcard will probably give better results than a cheapo, and the good ones may well be 24-bit while the cheap ones are 16-bit, but has anyone ever masked off the bottom 8 bits of a good 24-bit card and detected the difference? 73 Peter Hi Peter, Have you looked at how many bits are noise free on a good 16 bit card? 73 de Phil N8VB ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] 16 versus 24 bit audio
Tests done by Leif SM5BSZ have demonstrated that the Delta 44, one of the best sound cards around, when used in its 24-bit mode, gives about 18 usable bits of amplitude resolution. The rest is noise. I am sure that if the same test were performed on a 16-bit card, that would give certainly less than 16 bits of resolutions. So the answer is, yes, there is a point in using a 24-bit card instead of 16-bit one. Only do not expect to have a dynamic range of 144 dB... 73 Alberto I2PHD - Peter Martinez wrote: From G3PLX: I am just starting to do some serious tests with my new 24-bit Firebox. With the audio input of a channel short-circuit, if I analyse the statistics of the 24-bit data, I find the bottom 8 bits are ALWAYS random. In fact the bottom 9-10 bits are random. This tells me there is no point at all in having 24-bit audio. I might just as well stay with 16-bits. Has anyone here done any rigorous tests comparing 16 and 24-bit audio and found any difference? Like patching the code to mask off the bottom 8 bits completely, or replacing the bottom 8 bits with random data. Maybe someone has added a checkbox to turn this sort of test feature on and off and see what, if any, is the difference in the speaker. 73 Peter ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://mail.flex-radio.biz/pipermail/flexradio_flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com