Eleberoth2 - let me follow on this.
In point 3 he mentiones up to 2V of 50kHz frequency on the speaker cables - complete nonsenese - unless he talks about normal audio signal (my Rowland ha -3dB at 65kHz).
Point 4.
"Make the output impedance, a passive filter, variable with frequency and dependant on speaker loading."
I design electronics for 30 years but have hard time to understand how passive filter is variable with frequency. There are common mode chokes and capacitors on the output (zobel network) but they are present in most of other amps. This filter is set to about 65kHz (-3dB). Filter is within negative feedback and damping factor for Icepowe is about 4000. Early amps on Tact modules had output filter outside of feedback - but he makes general statements.
Point 5.
"Allow the amplifier to be marginally or completely unstable with high or open circuit output loading"
Complete nonsense - since switching amp is constantly unstable (cannot become more unstable - big advantage) being analog modulator/oscillator.
Point 6.
"Employ soft compressor clipping circuits prior to full power clipping"
Yes it does - (big advantage) designed in on purpose to protect tweeters from high energy during overdriving input. Many amps (like NAD) have this advantage
Point6.
"Employ high order negative feedback to improve in-band distortion figures"
Every amplifier does this either with global or bunch of local feedbacks. I know about multiple feedbacks in Icepower but suspect them to be shallow since output is pretty linear. I don't know how he knows about high order but Class D is different creature and has different feedbacks. He might be implying deep negative feeback - usually bad because of TIM distortions but there is no need for deep feedback and even if they put one in it cannot cause TIM. TIM comes when feedback cannot react fast enough and output transistors go to saturation getting charge trapped at their junctions becoming immune to following signal for a while. In Class D output transistors work differently (switching) and are respodning to time not voltage.
Point 7.
"Use feedback to provide numerically high damping factor at low frequencies and claim that this guarantees fine bass.(regardless of the interface to the loudspeaker)"
Isn't he implying by word "numerically" that this damping factor is not real and manufacturer is not honest. That is what I call manipulation. He makes impression of poor bass performance of Class D - this won't fly either. Enyone who listened to Class D amp knows tahat bass performance is its best feature.
Speaking od damping factor - my Icepower has 4000 and probably unnessesarily since inductor in series with woofers has obout 80mOhms and limits DF to 100. At 20kHz similar limitation is imposed by about 0.2uH imductance of the speaker cable. DF of 200 would be plenty. |
Guidocorona - Hi-Fi critic is wrong on all 17 points he presents. Class D produces only about 1% of the carrier (usually 0.5MHz) on the speaker and practically no switching noise (zobel network). My Rowland 102 is directly under TV - no problem. At 0.5MHz speaker cable needs to be 150m long to be 1/4 wave antena. Class D Amps cannot be "unstable" (big advantage) because by definition they oscillate (analog modulator). They are in reality high power sigma delta modulators. It is important to realize that most of modern dacs (sigma-delta) are working on principles of class D (PWM) - same with SACD (PWM at 2.8MHz). I read a little bit of Karsten Nielsen doctorate work on Icepower (University of Denmark) - very interesting. |
If Martin Collums does not like the sound of switching amps that's his right, but his 17 points are garbage. He thinks that switching amps are digital and have limited resolution (like in 60's). He claims that switching amps have "fragile output devices" where in reality they are more robust than class A/AB amps. For instance output of Icepower contains H-Bridge of 4 very strong Mosfets with all sorts of protection. He writes nonsense about power supply folbacks and heat dissipated in output devices. He appears to have no education in electronics and his statements are pure demagogy. His points about usage of negative feedback, high output impedance or "clipping" are all nonsense. He is right about soft clipping - B&O added it to protect tweerters from possible high energy during overdriving (Class D has a lot od power). He makes it sound bad but it is a virtue. Many traditional amps have soft clipping (NADs for instance). He mentions high intermodulation distortions where in reality it is just opposite - having no nonlinear output devices does not suffer from IM (Icepower 200ASC has 0.0005% IM at 10W 4Ohm 14kHz/15kHz). I could discuss every single point he makes - all of them are garbage. It is much better to say "I don't like the sound" instead of making uneducated false claims. |
No I am not from Denmark. Nielsens doctorate is available on internet. I own Rowland 102. At first sound was a little "lean" and sharp but over time it mellowed. It needs a little more "chestiness" (lower midrange) - improved after switching speaker cables to Acoustic Zen Satori. Transparency and Clarity is great. Bass performance is outstanding. Stereophile, 6Moons and others had glowing reviews of different class D amps - also Channel Islands D100 (based on Hypex modules - and no I am not Dutch either). The very best class D amps are probably Kharma (not Icepower) and Red Dragon (both too expensive for me). Sound is a matter of taste (so to speak). There is no rights or wrongs. Tube amps producing/exaggerate even harmonics sound wonderful on guitar or voice but horrible on instruments with more complex harmonic structure like piano. Piano structure is a little like percussion instruments that not follow simple harmonic structure and often sounds "out of tune" on warm amps. I prefer neutral amps. The question also should be "what is best for the money" - class D is difficult to beat. |
There are 3 kinds of switching noise in Class D amp:
Carrier frequency - approx 0.5MHz
Switching spikes of output Mosfets - approx few ns
Switching power supply frequency approx 50kHz
Carrier has very little chance to get thru linear power supply with all capacitors then power transformer and finaly input chokes and capacitors (pi filter). It has even less chance to get thru regulated switching power supply (my Rowland).
Switching spikes have very little energy and are filtered on power supply as above, and on the speaker cables by means of common mode chokes and capacitors (zobel network).
Carrier in not completely filtered on speaker cables (about 1% left) but needs 525 fett of speaker cable to become 1/4 wave antena.
Switching power supplies are present in probably every modern TV set and are strictly regulated. Not only that my Rowland for instance has CE certification (part of which is measurement of radio emissions) but also Icpower module used inside (200ASC) has folowing info in the datasheet:
EMI conforms to: EN55013
EN55020
EN61000-2
EN61000-3
FCC part 15-B
Elberoth2 - There are few early amps with poorer filtering like previous EVOs (not Icepowers) from Bel Canto or some NuForce amps but he implies they have poor performance because of it or that the company performs some illegal acts designing the switcher.
His point 1. is practically same as ponit 2. It is not possible to push noise out without pushing in - it's the same thing. He just makes it sound worse. |
Elberoth2 - jus shortly
9. "Use steep low pass filters to limit the upper high frequency range, partially negating the purpose of wider bandwidth, e.g. SACD, source material while resulting filter phase shifts may be audible in the working band"
Don't know what he means - Icepower has 65kHz bandwidth (-3dB) and typical phase shift at 20kHz.
10. "Have input circuits which are highly susceptible to higher frequency input signals including upper band noise shaper noise and DAC artefacts, and which then contribute to poor treble sound"
Why input circuit are more susceptible ti higher freaquency - I don't know.
11. "Have output circuits with poor high frequency resolution resulting in high levels of intermodulation products at the high frequency end of the spectrum"
Resolution is unlimited (analog) and IM distortions for my Rowland are 0.0005% at 10W 4 Ohm 14kHz/15kHz. How many other amps can do that?
12. "Have 'sampler' noise shaped noise floors which vary dynamically with the level frequency and complexity of the input signals"
I don't understand this language - what samplers is he talking about in analog modulator?
13. "Have comparatively small power supply reservoirs, in the light of their low frequency output current potential"
My Icepower being smallest of the series has peak current 11A. Larger 201 has 20A and largest 501 has about 50A.
14. "Have thermal dissipation limitations due to the small power module size which means that thermal dynamic variations are present in the performance with time"
Nonsense - my amp dissipates abot 5W and runs completely cold. Output switchers (Mosfets) have close to zero resistance when "ON" and switch within nanoseconds - not much power dissipated there (even no heatsinks).
15. "To protect the fragile output stages all kinds of pre-clip, and aggressive fold back protection regimes are included which are frequency dependant and are also programmed for duty cycle"
Again - tossing terms. Output is more robust than traditional amps having strong Mosfets in H-Bridge and all sorts of protection. Foldback protection is used only in linear regulated supplies (not used in amps for output power). Someone was testing Icepower at full power with music, sinewave and noise for many hours - found them to be exceptional. I can find it if you want.
16. "Operate at an equivalent sample rate which is insufficient for good resolution above 7kHz. DSD 1 bit pulse-width modulation operates at 2.4MHz, nearly ten times the rate used in Class D amplifiers"
It does not sample anything and has unlimited resolution at any frequency (it is analog). Has bandwith limit of 65kHz because of 0.5MHz carrier. 2.4MHz he mentiones or 2.8 MHz with SACD has different purpose (sampled system Nyquist - there is no DAC here)
17. "Deliver high constant DC voltages relative to local ground (up to 70V) at the output terminals (of course not between the +,- terminals) and hence the loudspeaker connections and cable"
DC voltage on output of my amp is under 1mV. What he means by "High Constant DC Voltage"? I don't know terms like that. This is garbage - believe me |
Just one more comment to pnt. 17 - I said I cannot understand what he's saying. It looks like he things there is up to 70V of constant DC voltage on the speaker terminals???? Please help me to understand this!!
I don't know what bothers me more claim of 70V DC on the speaker terminals or terms like "Constant DC" (had impresion that DC is constant by definition) |
Muralman1
ICE modules do not sample - they have analog sinewave oscillator/modulator converting input voltage to duty cycle and drive output switches (Mosfets). They resemble now more of sigma-delta modulator.
Hypex is also analog and probably most of the others.
(Hypex has half bridge output and sounds more like very good Class AB amp versus ICEPower with Full Bridge sounding more like very good tube amp)
Sampling is not even possible since to preserve 16bit resolution at for instance 1kHz - carrier frequency would have to be 65536*1kHz=65MHz. Just imagine bandwidth of my Rowland would require 65kHz*65536=4300MHz carrier. I wish it could be possible since one could take SPIDIF and run to PWM using just some DSP manipulations.
I heard claims of small amount of carrier residue causing intermodulation with audioband on the tweeter - which is only possible if tweeter can move membrane at this (0.5MHz)frequency.
Oversampling itself was created to allow filtering of the carrier with gentle filters with even group delays (Bessell) preventing wrong summing of harmonics in passband. My Benchmark DAC1 uses statistical manipulation to achieve equivalent of 1milion times oversampling and 24-bit resolution to drive output DAC at only 100kHz (audio DACs have lowest THD at around 100kHz).
Stereophile reviewed CODA S5 class A amplifier and compared findings to older Class D Bel Canto Evo amplifier. CODA was little bit better. In letter to Stereophile president od CODA expressed happiness that his amp came favorable in comparison with as he said "such great amplifier as Bel Canto Evo". There is no more honest praise of the product than one made by competition.
Interestingly Coda had more hiss at the tweeter than Bel Canto.
I chose Rowland since I heard That Jeff Rowland is classy guy and did not released bad or mediocre product in 20 years. He switched now his production to 100% Class D (Icepower). Rowland 102 is built nicer than Bel Canto and was cheaper at the time. I have heard a lot of very positive things about H2O. |
Atmasphere
Peak current I mentioned is the current limit of the amplifier output to speaker. 20A means that my amp can put that much current on the speaker. Nothing unusual - many amps have similar spects. It is rated 100W at 8 Ohm and 200W at 4 OHm - both with 0.2% THD 10Hz-20kHz. As for impedance - amp can drive 3 Ohms min (larger ICE drives under 2 OHms). |
Branimi
I'm not from Italy either - cannot understand a word.
My Icepower (200ASC) has DF=30 at 10kHz (nothing unusual about it). At 1khz DF=1000 and climbs to 4000 at 10Hz. |
Ar-t
I agree absolutely. Synergy is probably one of the most important things.
I respect people who can honestly say "I don't like it" even if it costs $500k, but I don't care much for blindsided people who try to present pseudo-sientific reasoning.
Being subscriber to Stereophile for many years I have a lot of respect for technical knowledge and integity of people there. They present detailed measurements but put more stress on impression with different co-equipment and different music pieces. Technical aspect is always interesting but is secondary to sonics. |
Simontju
Amp is close to 95% efficient and does not have to be large to do so.
"thermal dynamic variations are present in performance" - thermal dynamic variations of what? This is still jibrish to me.
Let me place here results of test made by somebody to 1kW icepower and 100W Icepower (same as mine). Notice that he tests 1kW amp with no hestsinks at all. If you take into consideration that average power in music i very low (only few percent) statements about "Thermal dynamic variations" are complete nonsense. Read below:
"Originally posted by dmfraser
I operated a sample 1000ASP on the bench delivering 350W average of pink noise into a 4 ohm load for over one hour with no additional heatsink and the metal case stayed below 55°C.
However, higher levels would make the power supply voltage go down to act as a thermal compensation. Much nicer than just shutting down.
However, with 1214W of sine wave, partly into clipping, after about 35 seconds, the output level would drop to about 600W by the protection circuitry. Remember this is with no additional heatsink.
The 200ASC, I was able to operate with sine wave at 100W into 4 ohms continuously. Turning it up, with sine wave, the module would shut down at about 214W.
However, with pink noise at 100Waverage and a 6 db crest factor, the 200ASC would run all day without shut down, until I was tired of tying up one of my Audio Precisions with it. It would also run all day with music, driven hard where the output limiter was set to a peak limit of 210W maximum. That is, the limiter would allow small amounts of clipping. Driven so the limit light in our external optical limiter was on essentially continuously.
We're quite satisfied and will be ordering production quantities of the 200ASC shortly." |
Atmasphere
Quoted current - as I said before is the peak current and has nothing to do with average power. It is used to show how well amplifier can respond to impulse.
Internally my 200ASC Icepower can deliver 48V to speaker what would make at 3 Ohm 16A current. Limiting circuitry is set to 11A. Power delivered to speaker will be 363W.
I don't know how long this pulse can be or what would be crest factor with multiple pulses but would like to point that Bel Canto rates the same module at 300W (10% thd) |
Muralman - I have no experience with high end DAC circuits but my Benchmark has surprisingly clean and transparaent sound. Part of this is due to anti-jitter circuitry. Jitter produces sidebands of very low amplitude (typicaly -80dB) but very audible since not harmonically related to root signal. It sounds less clean (hash) but often people prefare sound with some amount of hash or thd. They often call clean systems "analytical". Sombody even said, after listening to hi-end system, that everything (instruments) sound separate and he preferes all mixed together. I suspect I have similar hearing limitations with my audio experience.
I will search for more info on oversampling. I just don't know how they can filter out carrier of 44.kHz. With the ratio of 2:1 Bessel filters just start attenuating - no matter how many poles. Usage of other filters will affect the sound. |
Muralman - people are often confused and becuse they like guitar sound on 30W Vox (remeber The Beatles)with tubes and no feedback (high thd) and want their home system to be like that - close to real performance. Home system of course should not change sound - even if clean jazz guitar does not sound as lively as distorted one. Issue of real life performance is very interesting because real life performance is often very poor (poor acoustics, bad seat etc) - not the studio level what becomes clear after you compare sound from the studio and the concert records.
Jitter is resolved by means of PLL - single loop in cheaper and double in more expensive CD players. DAC spits out with the rate from the disc (to be exact from internal clock phase lock looped to CD and going thru FIFO buffer at this rate). Good transport means lower jitter but is costly. Another method is to completely reclock signal to DAC's clock, upsample it and send to DAC in corrected intervals. Benchmark is trying to do it (correct)acurrate to 5ps. Both methods produce great results - reclocking gives better supression of the jitter and allows to use cheap DVD player as a transport.
I checked audio DACs at Analog Devices website - almost all of them are sigma-Delta (oversampled). There is still few with traditional architecture carrently being discontinued. All Dacs shown as 24-bit are sigma-delta. Traditional Dacs go only to 18-bit (maybe 20-bit).
Dielectric in the speaker wire is evil (even teflon) - I agree. Latest trick is to use foams (lower dielectric constant than solid material) and wide tubes so that wire has only contact with one wall. Another problem is to compromise between low resistance and low inductance (thick wire) and skineffect free (thin wire). Skin effect starts at gauge 20 at 20kHz and splitting wire into many helps only a little (more surface area but still in the sum of magnetic fields). Tapes are wonderful but not practicall solution. I replaced coils in my cheap speakers with the "tape coils". It sounds better - at least I'm telling myself. Capacitor in series with the tweeter is the reson for brigtness in many speakers. Mylar caps are horrible (high dielectric constant) but Teflon are too expensive. Polypropylene paralleled by small teflon is the ticked. I also changed resitors to non-inductive. Every bit helps. |
Ar_t
Icepower uses full H-Bridge of 4 Mosfets switching direction of the speaker between +VCC and GND. In spite of VCC/2 on the speaker terminals, voltage between terminals is zero. The only disadvantage of this scheme I know is that you cannot bridge amplifiers. My amp has 23.5V on speaker terminals - it won't zap you (voltage up to 60V is considered safe)
As to effect on the damping ratio - filter is within negative feedback and DF is very high. At higher frequencies it drops as I stated before to about 30 at 10kHz. What about DF of a tube amp at 10kHz?
What about zobel networks on traditional transistor amps.
It has about 1% of the carrier - not important (cannot radiate and the tweeter produces IMD only if it's membrane moves - not possible at 500kHz).
Talking about "scare" - the scarriest part of Mr. Colloms presentation is the fact that he believes that amplifiers like Icepower or Hypex (he tested Channel Island D100 - didn't he) are sampling systems and have AD/DA inside with limited resolution.
Sure anyone can find something wrong with any type of amps. The wrong think to do is to pick possible deficiencies of different amplifiers and make general statement about whole class of them. At the same time he did not mention any positives like very low THD, extremely low IMD+TIM etc.
I would advise everybody to listen to Icepower, Hypex, NuForce, Flying Mole and make up your own mind. |
Ar-t - I have the smallest of the series and 23.5V on terminals. To deliver 5x more power they need voltage around 52.5V. I think it stays below safe level of 60V but agree with you that this might be of concern. I don't touch speaker connectors but with largest unit I would be concern about toddlers. Hypex is free from it.
As far as I know filter set at 65kHz has couple of chokes on one toroid (common mode chokes) and ouput capacitors. Frequency of this is in general fixed execept when Q is changing. Mine is listed at 65kHz at 8Ohm and 45kHz at 4 Ohm. Speaker impedance changes will affect frequency but it happens (if I remember) usually in lower frequency range. B&O lists idle frequency 400kHz-460kHz but for operating frequency they show 0-460kHz. I don't know what they mean - it cannot certainly be in audible range. Maybe they include audio signal in the spectrum?
So - there are some changes in the carrier frequency and therfore its attenuation. Audioband should have some phase shift - about 20 deg. increase at 20kHz compare to 1kHz. Is it common?
There is something different about highs of Icepower (might like it or not). Is it caused by limited bandwidth or dropping damping factor?
Some people called midrange of Rowlands 201 being magic and almost everybody likes bass performance but highs sound different.
There is no easy choice in design of amps. Class A has about 1/8 of efficiency and large power units are costly if not impossible. Class AB have nonlinear element in the output and require modesty in design. General problem is setting negative feddback to deep to get specs. THD for instance is difficult to hear below 1% but I would not buy class AB amp with 0.001% THD or very High DF. Sane designer would use the most linear elements setting bias current really high and would bring down thd only below 1% and then would limit input bandwidth to one amp had before feedback to prevent TIM. Unfortunately companies are racing for best specifications. I would advise to take specifications, promtly discart them and just listen.
Class D bypasses many problems and I know this is just begining. Comparing them to traditional amps is fair when is done to units in the same price range. |
