Messages

[THD + Noise spectra and RMS voltage of Channels-1/2 unbalanced output]

THD + Noise spectra and RMS voltage of Channels-1/2 unbalanced output

This thread is about improving the performance of an already great bit of test gear. If you don't have the necessary skills to do these mods or your instrument is still under manufacturers warranty then we don't suggest you perform these mods.

According to the specs the Prism dScopeIII has an equivalent input noise voltage of about 1.2uV which is equivalent to about -116dBu and which was also confirmed by NewAvGuy on his blog. We also confirmed this on our own dScope III by running a similar test by shorting the inputs and measuring the noise level over a 20kHz bandwidth and we also got close to -116 dBu which is only a few dB lower than the Audio-Precision analyzers which sell for a lot more than the dScope III



Is it possible to improve on this ? The ADC converter use in the dScopeIII is a AKM AK5394A which has a S/N 123dB so there should be some room for improvement. Removing the top cover of the dScope III reveals that all of the ADC and DAC circuitry is contained on a riser board that plugs into the switched attenuator board. On further observation we note some of the IC's are plugged into sockets which would indicate that the manufacturer was contemplating replacing the chips with better devices as they became available or simply they may have been trying different chips from a batch until they found the ones that met their specs. Bear in mind the design of this analyzer dates back to 2001 so it is quite remarkable that they were able to achieve these specs with the silicon they had at the time ! We note that the two IC's closest to the ADC are standard NE5532 dual opamps which would be part of the buffers/level-shifters used to drive the ADC. As these devices have an equivalent noise voltage of 5nV/rtHz it would be a simple exercise to replace these with devices with better noise an distortion performance. As high quality opamps with DIP packaging are slim pickings these days we ended using the tried and true LME49720 which is a drop in replacement with much improved characteristics over the venerable NE5532. The LME49720 has an equivalent noise voltage of 2.7nV/rtHz and an order of magnitude lower distortion so we could expect at most an improvement of noise reduction of 5.35dB but in reality it would be somewhat less. Lets see what happens



And from the new noise measurements we see an improvement of about 3dB from -116dBu down to -119dBu so well worth the few dollar investment of these opamps The question arises could we improve on this yet again ? Since the DIP style of package is becoming very rare for audio opamps, to get a lower noise floor in a dual opamp configuration would mean that we'd have to go to an SMD package and use some sort of SMD to DIP adapter so probably not worth it for chasing down a few extra dB's. Maybe an experiment for a rainy day but the current mod is quite easy to achieve for an hours work and a few extra dollars for the opamps

[Output level of 25dBU = 13.7 VRMS on Channels-1/2]

Output level of 25dBU = 13.7 VRMS on Channels-1/2

Note the differences between the two THD values is probably due to the measurements dwarfing the lower limits of the analyzer. (Anyone care to lend me an Analog-Precision APx555 ??)



Output level of 3 VRMS on Channels-1/2



Output level of 1.9 VRMS on Channels-1/2

[Noise spectra and RMS voltage of Channel-1/2 unbalanced output]

Noise spectra and RMS voltage of Channel-1/2 unbalanced output



Noise spectra above with expanded scale showing some very low level mains (50Hz) artifacts

[Noise spectra and RMS voltage of Channel-1/2 Balanced Output]

Noise spectra and RMS voltage of Channel-1/2 Balanced Output



Noise spectra above with expanded scale

[Test Results:]

Test Results:

• THD + Noise (Balanced Output)


• THD + Noise (Unbalanced Output)


• THD + Noise (Balanced Output - 80kHz bandwidth)

Conclusions:

The Ultimate-Preamp 2/+ has vanishing low levels of distortion and noise, so much so that it is approaching the limits of our dScope III audio analyzer and most likely exceeding it !! Even at much higher output voltages compared to your typical DAC the UP2/+ does not disappoint as shown by the following screen shots of the THD+N measurements 😁 The difference in distortion between the two channels is most likely due to the differences in the analyzer ADC and not the Preamp because reversing the channels does not appear to have any effect on this 😉 To get better measurements we would need access to an analyzer such as the Audio Precision APx555 which nulls out the fundamental so that it can more accurately read the harmonics from the output of the Preamp rather than the added distortion from its own ADC. However, we are not complaining about these results because it looks like all of the work we have done on the DAC and post DAC circuitry in this new Preamp has payed off

The THD + Noise performance of the unbalanced output is a whisker higher than the balanced output but still does not disappoint ! Below 1kHz one can clearly see the mains artifacts that are not present in the balanced output spectrum but are still buried way down below the threshold of hearing of -120dB and so of no concern !! These artifacts are caused by leakage flux from the transformer used in the linear power supply setting up earth loops in the internal wiring from the DSP board to the Unbalanced output board. Yes folks, balanced always wins out simply because common mode noise is cancelled out !! Above the 1kHz fundamental test signal we can see the usual harmonic distortion spikes mainly attributable to the distortion of the ADC used inside the analyzer. There are no spurious noise spikes from an external switch mode power suppliy simply because we don't use one and the on-board switching regulator used to power all of the digital circuitry runs at 500kHz which is well outside the audio bandwidth !

[Test Results:]

Test Results:

• Noise performance - Balanced Output
• Noise performance - Unbalanced Output
• Noise performance - Balanced Output - A-weighted
• Noise performance - Unbalanced Output - A-weighted

Conclusions:

The noise floor of the balanced output of the UP/2+ is only 6dB above the noise floor of the dScope III audio analyzer which is approaching -116dBu or 1.2uV  RMS ! Bear in mind that unlike standard DACs we have incorporated a buffer amplifier on each balanced output to boost the output level from the output of the IV converters which will of course add some additional noise of its own. Also in the 2nd image capture below we have expanded the frequency axis from 0-1kHz to show the almost non existent mains components sitting above the noise floor which is a tribute to the balanced design.

The noise performance of the unbalanced output is slightly higher than the balanced out but still does not disappoint ! On the expanded frequency scale one can clearly see the mains artifacts that are not present in the balanced spectrum but are still buried way down below 120dB !! This is caused by leakage flux from the transformer setting up earth loops in the internal wiring from the DSP board to the Unbalanced output board. Yes folks balanced always wins out simply because common mode noise is cancelled out !!

dScope III Noise Performance:

• Unmodified dScope III Noise with inputs shorted

• Upgraded dScope III Noise with inputs shorted

• Upgraded dScope III Noise with inputs shorted (A-weighted)

[References:]

Measurements are done with a variety of instruments namely :-

• Prism dScope III Audio Analyzer,
• Agilent 34401A 6.5 digit Precision Digital Multi-meter
• Tektronix Digital Oscilloscope models, TDS784A, TDS3052, TDS7054
• Tektronix CRO model 2246A
• Tektronix Arbitrary Function Generator model AFG310
• Hewlett Packard AC RMS Voltmeter model 3400A
• Leader LDC-824 6.5 digit 500MHz Frequency Counter with Oven Controlled crystal oscillator

References:

• dScope III Specifications


• Handy online Conversion Calculator


• A description of measurements and procedures by Amir from ASR


• A description of measurements and procedures by Amir from ASR


• Test Methods and Comparison between measurement equipment by the NewAVGuy

[Hypex NC252MP THD+N vs Power @ 1kHz into 4 ohms, Single Channel Driven]

And now for the Hypex NC252MP Ncore power amp measurements. The Ultimate amplifier uses two of these modules to give 4 channels of additional amplification. The name plate rating of these amps are 250 and 200 watts into 4 and 8 ohms respectively @1% THD. We measured the power to be 306 and 159 watts into 4 ohms and 8 ohms respectively at 1kHz and 1% THD. Whilst the 8 ohm power measurement is down on the manufacturers name plate rating the doubling down of power into 4 ohms shows how well regulated the switch mode power supplies are. 

Hypex NC252MP THD+N vs Power @ 1kHz into 4 ohms, Single Channel Driven



Hypex NC252MP Power at 1%THD @ 1kHz into 4 ohms, Single Channel Driven



Hypex NC252MP THD+N vs Power @ 1kHz into 8 ohms, Single Channel Driven



Hypex NC252MP Power at 1%THD @ 1kHz into 8 ohms, Single Channel Driven

[Hypex NC502MP THD+N vs Power @ 1kHz into 4 ohms, Single Channel Driven]

The Hypex NC502MP is the most powerful amplifier used in the Ultimate Amplifier and represents the first two channels (1-2) out of the 8-channels of amplification. Typically it would be used to drive the woofers in a stereo active speaker system so it is imperative that it does this without any problems. With a name plate power rating of at least 500 and 350 watts per channel into 4 and 8 ohms respectively the measurements we made did not disappoint. The following measurements were made using the dScope III by sweeping the power from 10mW to beyond clipping and plotting the THD at 1kHz. The maximum power output is measured at 1% THD @ 1kHz the same as in the Hypex data sheet. The results show that the 4 ohm single channel measurements of 750 watts @ 1% THD showed that the manufacturers power rating is very conservative for these modules. Also there is doubling of power from 8 ohms to 4 ohms which shows that the switch-mode power supplies are extremely well regulated which is a testament to whoever designed the power supplies ! The low actual distortion measurements also confirm the measurements presented in the manufacturers datasheets for this module

Hypex NC502MP THD+N vs Power @ 1kHz into 4 ohms, Single Channel Driven



Hypex NC502MP Power at 1%THD @ 1kHz into 4 ohms, Single Channel Driven



Hypex NC502MP THD+N vs Power @ 1kHz into 8 ohms, Single Channel Driven



Hypex NC502MP Power at 1%THD @ 1kHz into 8 ohms, Single Channel Driven

[The first incarnation resulted in a 1dB drop at 20KHz]

To measure class-D amps using an audio analyzer with digital processing it is prudent to filter out any ultrasonic carrier noise (300kHz and above) as this can fold back into the audio spectrum and interfere with the measurements by exaggerating the noise measurements even though the noise is not part of the audible spectrum and for all intents and purposes is essentially filtered out by the speakers and certainly cannot be heard by the ear !.

Both Prism Measurements and Audio Precision sell outboard filters designed for this very purpose however we decided to build our own based on Prism's own specs which is very similar to a Chebychev 3rd order low pass filter with a 0.05dB equi-ripple response in audio pass-band and at least 50 dB attenuation at 200kHz.

http://www.prismsound.com/test_measure/products_subs/filter/lpf_spec.php



Our filter design yields a very similar response



The first incarnation resulted in a 1dB drop at 20KHz



For some reason the calculated values for the components was yielding an attenuation of nearly 1dB at 20KHz but with some experimentation with the simulator a better match was obtained and now the filter is within 0.05dB from 20-20KHz

Measurements are done with a variety of instruments namely :-

• Prism dScope III Audio Analyzer,
• Custom designed LPF for measurement of class-D amplifiers similar to Prism ds-LPF
• Agilent 34401A 6.5 digit Digital Multi-meter
• Tektronix Digital Oscilloscope models, TDS784A, TDS3012, TDS7054
• Tektronix CRO model 2246A
• Leader LDC-824 6.5 digit 500MHz Frequency Counter with Oven Controlled crystal oscillator
• E-MU 0404 external 24 bit 192KHz sound card

Although one of the benefits of class-D amplifiers that has been touted for so long is efficiency, in practice the amps can still generate a sufficient amount of heat and the switch-mode power supplies also contribute to this heat. The Hypex NCore amps are no exception. A look at the data sheet shows that for the NC502MP producing 400 watts a channel into 4 ohms the amps will dissipate about 200 watts which equates to around 80% efficiency. Whilst this looks good on paper, 200 watts of power dissipation will quickly raise the temperature of the small plate heat-sink on the amp module to unsafe levels unless this heat can be channeled away quickly. Then there is the question of dealing with 3 more stereo amps for a total channel count of eight amps each dissipating a given amount of power depending on the frequency response and program content. Although the power output of the amps should diminish the higher the frequency range they are ask to handle, the thermal requirements should not be assumed to be negligible. This a mistake many people make with class-D amps by mounting them on a thread-bare chassis with minimal thermal capacity thus questioning the long-term reliability of the design as a whole.

Our custom designed case works exceptionally well at dissipating huge amounts of thermal energy away from the amp. All of the amps are mounted on a 10 mm thick aluminium heat spreading plate with a large surface area the size of the bottom of the case. This acts as a transient thermal sink with a huge thermal capacity and which can sink a huge amount of transient energy without raising the temperature quickly to unsafe levels. Then the wrap around case with a large amount of surface area acts to slowly dissipate away this energy using convection.

Add to this is our thermal management system which monitors and reports temperature extremes on a channel by channel basis whilst maintaining temperatures inside the case at safe levels.  We deploy a 60mm fan chosen for its low noise level along with a dedicated fan controller which employ both a fan and thermal closed loop control system which carefully ramps up the fan speed over a range of temperatures in order to keep the temperature of the system at a sustainable level and without generating intrusive noise.

The firmware has a diagnostic mode which can continually report the amplifier status over a serial port, including temperatures and fan speed. We can see the first two channels are working hard because the on-board thermistors are reporting higher temperatures than the other channels which are sitting idle. The last two channels are not showing anything because the HF100 modules do not include the thermistors. The temperature of the heat spreading plate has risen to 34 degrees and the fan has already spun up to 1500 RPM to get rid of any excess heat inside the case.  If the thermistor temperature rises to 80 degrees C the respective channel indicator LED will turn white. Further increases above 90 degrees will shut down the amplifier until the temperature is reduced to 80 degrees.

[attachment id=1 msg=20]

[Scope image of the first two channels of the amplifier being driven into voltage clipping.]

Testing the clipping indicators. The Hypex amplifier modules can assert their channel clipping flag if the output of the amplifier exceeds either a set voltage or current limit. The clipping flag from the Hypex amplifiers are non-latched which means we need to latch the clipping flags in hardware otherwise the software could miss it if it is relying on polling. As a result each channel can report the clipping status in real time by making the LED turn red for at least 100mS so it can't possibly be missed by the user. This tells the user that amplifier is being over driven and the level should be reduced !

Scope image of the first two channels of the amplifier being driven into voltage clipping.




Indicator LEDs turn red when its respective channel is being over-driven into clipping thus signalling to the user to reduce the level !

[Continuous 1KHz Sinewave into 4 ohms before clipping with one channel driven corresponding to 595 watts into 4 ohms.]

Some simple power measurements on the Ultimate Amplifier using a scope. The first two channels are powered by a Hypex NCore NC502MP amplifier module which would typically be assigned to the low frequency channels in an active speaker system. This is the most powerful of all amplifiers in the Ultimate Amplifier with a name plate rating of 500 watts per channel into 4 ohms. We used the channel clipping indicators to determine the maximum power threshold before clipping but later on we will use the dScope III audio analyzer to accurately measure the power output at a given THD level and show how conservatively rated these amps really are The results will shock you at how powerful these amps really are and how well designed the switch mode power supplies are to maintain tight regulation and almost doubling power into 4 ohms

Continuous 1KHz Sinewave into 4 ohms before clipping with one channel driven corresponding to 595 watts into 4 ohms.





Continuous 1KHz Sinewave into 8 ohms before clipping with both channels driven corresponding to 321 and 317 watts respectively.





Full Power 20Hz Tone-burst with 1 to 5 duty-cycle (5 cycle on 20 cycle off) single channel into 4 ohms





Full Power 100Hz Tone-burst with 1 to 5 duty-cycle (5 cycle on 20 cycle off) single channel into 4 ohms





Full Power 1000Hz Tone-burst with 1 to 5 duty-cycle (5 cycle on 20 cycle off) single channel into 4 ohms





Top Trace = Full Power 100Hz Tone-burst with 1 to 5 duty cycle with 5 cycle on 20 cycle off single channel into 4 ohms
Bottom Trace = Current measurement of voltage across a 0.1 ohm sense resistor (20 amps per division)





Full Power 1000Hz Tone-burst with 1 to 5 duty-cycle (5 cycle on 20 cycle off) single channel into 8 ohms





Continuous Sinewave at 1000Hz After Clipping with both channels driven into 8 ohms