Our first HDMI Benchmark article generated a lot of feedback and support from our readers, but also a lot of skepticism. (Read Part 1 of the HDMI Benchmark)

The most common thing we heard from skeptics was a form of “I can’t see a difference between players” or “Benchmark numbers don’t mean anything if you can’t see it in the real world”. One of the reasons that we used CIE1994 dE numbers for our results is that the CIE1994 formula is based on what the human eye can see. Numbers below the 1-1.5 range are differences that are not visible, and beyond that they are.

We had another idea for a test that we thought would be of interest to our readers as well, which we will present today.

The Tests

While many people do hire a certified calibrator to come and setup their display, far more people rely on the many DIY options that are available now. For almost all of these, your Blu-ray player is the source used as it’s cheap, or free, to acquire a calibration disc for one compared to buying a dedicated signal generator, or hiring someone to come out. However, if your Blu-ray players output is incorrect, how would this affect your actual calibration? To test that we used the following equipment:

  • A Lumagen Radiance Mini3D Video Processor
  • A Blu-ray player with correct output, we will call it Player A
  • A Blu-ray player with incorrect output, we will call it Player B
  • A calibrated i1 Display2 colorimeter
  • ChromaPure version 2.2 Software
  • The AVS Rec. 709 Calibration Disc
  • A Samsung PN50B650 Plasma Display

We wanted to examine this two ways: Using an incorrect source on a correctly calibrated display, and using an incorrect source to calibrate a display. The methodology for Test 1 was as follows:

  • Reset an HDMI Input on the display to factory settings
  • Let the display and colorimeter warm-up
  • Use ChromaPure with the Lumagen Radiance to perform an automatic calibration
  • Record the measurements of the display using the Lumagens internal pattern generator
  • Record the measurements of the display using the Rec. 709 Calibration Disc and both Player A and Player B

We made sure that both players were at factory settings, using 4:2:2 YCbCr output at 1080p60 over HDMI. 24p mode was disabled on the players as well as the display (24p mode can cause a change in black levels on many displays), and both players used the same HDMI cable into HDMI Input 1 on the Lumagen Radiance. This should ensure that the results are fair and only affected by the performance of the Blu-ray players themselves.

For Test 2, the methodology was:

  • Reset an HDMI Input on the display to factory settings
  • Let the display and colorimeter warm-up
  • Using the Player B and the Rec. 709 Calibration Disc, calibrate the display using its internal CMS and ChromaPure
  • Record the measurements of the display using ChromaPure
  • Using the pattern generator of the Lumagen Radiance on this newly calibrated input, record measurements in ChromaPure from the display

The same precautions were used as above, with the same HDMI cable for both, and nothing else in the signal path. If Player B really is not causing any real world issues, than we would expect the calibration results to be identical.

Test 1 Summary and Results

Using the Auto-Calibration feature of ChromaPure and the Lumagen Radiance, you can see that the display now has fairly good grayscale tracking, near perfect gamma, and very accurate color decoding. Other than an issue with Blue, all errors are below what should be easily visible. When we run the calibration report using Player A, we can see that the measurements are very close to those of the Lumagen Radiance. The difference in them is small enough to easily be meter error, and had I used a setting of 10 instead of 3 in measurement smoothing they might have been even closer, but it would have been a much longer testing period.

When we look at Player B, we can see three main issues: The Gamma is now much higher than it should be, the peak light output (Y for 100% white) is much lower, and the dE of the grayscale is higher. The most noticeable change is the Gamma value, which is almost off the chart. What is happening is that Player B is incorrectly adjusting the Luminance value of the data, and pushing what should be an accurate Gamma up to a much higher value, which will result in a much darker, flatter image overall. The other issue is that the peak white value has been reduced, so that the brightest whites won’t be as bright as they should be.

This incorrect Luminance also shows up in the color decoding, as Luminance (Y in YCbCr) is incorporated into all the colors when it is decoded into RGB. If the Luminance is off, the value of the RGB values is going to be incorrect, and we can see that in these charts. Player A and the Lumagen are nearly identical as colors go, but Player B shows some significant differences in color luminance, particularly green and magenta. While not as bad as the gamma, it certainly is not correct and something that you could possibly notice on a correctly calibrated display.

In the end, Test 1 shows that if you use a Blu-ray player with incorrect colorspace conversion on a correctly calibrated display, you can have some serious image issues that will crop up and be visible.

Test 2 Summary and Results

For this test, we calibrated the display using Player B, and then measured it using the Lumagen Radiance to see what our display is actually calibrated to. In the charts, the values marked as Player B are what we measure when using Player B as our test pattern source, and Actual is what we measure when using the Lumagen Radiance as our test pattern source.

Using the internal CMS of my display, you can see that we appear to have actually managed to get a very good calibration done. The grayscale error is down to nearly nothing and the RGB tracking is far better than before. At least we think we have a good calibration, so we will use the Lumagen and see what the actual readings from the display are when using a correct source and how those now compare to our target values.

This time we notice that our gamma value is dropping down a lot from what we thought it was to something much lower. The dE for our grayscale is going up across the board, and our peak luminance value is much higher than expected. Looking at the color errors we see that while our blue error is lower than expected, all of our other colors have higher dE values than we thought they did.

The real world implications of this would be that any sources using this display would lose a lot of their contrast due to the incorrect gamma curve, we could suffer from blown-out highlights as our peak white value is coming out much brighter than expected, our grayscale is not as accurate as expected, and our colors show more error than expected. By calibrating this display to account for the incorrect decoding of Player B, we have messed up the settings for all other components that would correctly display.

The biggest sin is really the incorrect gamma, which will cause your image to lose a lot of it’s pop, no matter if it’s color or black and white, as it will flatten out and dull the range of contrast you can expect in your display.


In the end, there really is no way to calibrate around these errors, as hooking the player up to a properly calibrated display will result in a dark image and missing highlights in this case, or you can calibrate your display for the player, and then suffer blown-out highlights and a lack of contrast and pop in your image for all other sources. Of course, the issues vary player-to-player, but the end result is the same: compromised performance.

Even when the decoding and conversion errors in your player are small, those can add up as the signal moves through the rest of your signal chain. Your display will certainly have some error in it, and perhaps your receiver or processor will introduce errors as well. All of these small amounts can wind up producing an error large enough for anyone to notice. Some people may still not care that their player has problems in its output, but since there are players out there that decode and output to all colorspaces correctly, there is no reason we have to accept this. We hope this helps to further explain our new testing methodology.

Harman Kardon HK 990 Stereo Integrated Amplifier with Digital Room Correction and Dual Subwoofer Bass Management – Part III

Circuit design and the tape recorder section are the focus of the final section of my HK 990 review. The topics are addressed to different groups. Those interested in the tape recorder section will find information on Page 9. The tape recorder interface section outlines matters of connectivity and usage.

Introduction to Harman Kardon HK 990 Stereo Integrated Amplifier with Digital Room Correction and Dual Subwoofer Bass Management – Part III

Circuit design and the tape recorder section are the focus of the final section of my HK 990 review. The topics are addressed to different groups. Those interested in the tape recorder section will find information on Page 9. The tape recorder interface section outlines matters of connectivity and usage. Usability issues are discussed. The tape recorder interface section is at the level of the prior two parts and assumes no knowledge of circuit-level electronics.

The circuit-design section enters the land of high-end design, mostly guided by the design principles of Matti Otala. My intent here is two-fold: to identify anything that might make the unit sound better subjectively, and to objectively quantify whether any aspects of the design have inadvertently degraded performance.


    • Design: Solid State Stereo Integrated Amplifier
    • Power: 2 x 150 watts RMS into 8 ohms @ 20 Hz – 20 kHz, 2 x 300 Watts into 4 Ohms
    • MFR: 10 Hz – 100 kHz
    • THD: <0.07% at Full Output (8 Ohm Load)
    • Analog Inputs: 7, Plus 1 Phono MC, 1 Phono MM, and 1 Balanced XLR
    • Digital Inputs: 1 HRS-Link, 2 Optical Digital, 2 Coaxial Digital
    • Analog Input Sensitivity/Impedance: 350mV/43k ohms for tuner/CD, 10mV/47k ohms for Phono-MM, 1mV/100k ohms for Phono-MC
    • Digital Input Capability: All Standard Digital Formats
    • Dimensions: 6.4″ H x 17.3″ W x 17.5″ D
    • Weight: 43.2 Pounds
    • MSRP: $2,599 USA
    • Harman Kardon
    • SECRETS Tags: Harman Kardon, HK 990

This circuit design section assumes the reader has knowledge of analog design equivalent to a 1980’s Audio magazine. I started reviewing in the 1980s. I took my cues at what level to write from Audio and never let go. Some of the technical terms bubbled to the surface in the HK 990 literature as the marketing folks tried to capture the unique aspects of the unit. You get only words without graphs or block diagrams. Usually, only when engineers write the literature does it all jell. Accuphase is an example of this practice operating to perfection. The literature Harman produced in the 80s and early 90s included circuits, novel measurements and even Bode plots. Sansui and Kenwood were producing similar material. Sansui was a big loss from the creative circuit view when it went out of business.

Here are links to Part I and Part II of this series.

Construction of the Analog Blocks

A textbook by Bob Cordell is an excellent reference. The book is oriented to those interested in audio-electronics design.

During the early 90’s, I summarized basic fundamentals in several issues of The Audio Critic and analyzed fully-designed commercial components as case studies. Issues 18 and 20, focus on preamplifiers and power amplifiers, respectively.

Free PDFs of these articles are available at

For those with a limited knowledge of electronics, the classic The Art of Electronics (Horowitz and Hill) is a must-have. Though pricey and heavy (at 1100 pages), the breadth of material well serves the needs of both novices and experts. If you are lucky, it may be in your local library; otherwise, try to secure a copy through interlibrary loan.

Significantly cheaper is the 5th edition (2011) of Teach Yourself Electricity and Electronics written by Stan Gibilisco. I have not reviewed the text, but it has been recommended by others whose opinions I value.

HK 990 Volume Control

The volume control is a good starting point because it is the one spot in the HK 990 where Otala design concepts are violated. Two stages of operational amplifiers are present when an analog source, including phono, is selected.

The Analog Device AD825 opamp is used. This is not a discrete circuit, which is surprising. A discrete circuit allows the amplifier stage to be optimized to all Otala requirements. A signal coming from digital media often sees many opamps in the recording studio, but this is not true for analog signals entering the HK 990 or signals from the phono preamp. It is especially surprising given that a discrete circuit is at the output of the HD 990 CD player.

The AD825 has some unusual characteristics that are more consistent with Otala design rules. That said, the opamp designers were probably unaware of these rules and coincidentally came upon them during optimization of the opamp for applications outside audio. The AD825 in the unity-gain configuration has a return-loop gain of 2000 (66dB) from 20Hz to 10 KHz. It then declines at 6dB per octave. A typical opamp has an open-loop gain of 100,000 to 1 million and will start rolling off around 10 – 100Hz. The AD825 has a high slew rate (115V/us), wide bandwidth (34MHz as a unity gain buffer), and sizeable open-loop linear input range. Indeed, Walt Jung references an Otala Audio Engineering Society (AES) paper in his Electronic Design article of  December 1, 1994 on the AD825. Still, it is not fully complementary like other Harman discrete circuits and methods that that prevent oscillations when the feedback loop is closed are different.

The downsides of the AD825 include higher noise levels than the typical audio opamp and some discrete designs.

The digitally-controlled volume IC in the HK 990 is a JRC NJW1159 comprised of only silicon switches and a resister ladder. There is no operational amplifier on the chip. With a specified maximum supply voltage of maximum +/- 7V (limited by breakdown voltages of the switches in the IC fabrication process used), the JRC NJW1159 requires a pair of dedicated sub-voltage regulators.

The 50k ohm input impedance of the JRC NJW1159 is typical of an analog control. However, MOS transistors are sensitive to electrostatic discharge and stress when the input voltage exceeds the +/-7V power supplies. The AD825 unity gain buffer in the signal path protects the JRC part. A DC blocking capacitor is at the input of the buffer. Another is at the input of the volume control. A second AD825 with a gain of 2.4 (8dB) at the output of the volume control prevents it from being loaded by the preamp output jacks. I calculated the input impedance of the power amp at a low 10k ohm, meaning the second AD825 is required for the direct power amp connection. The power amp has an above-average gain of 32dB to compensate for the low gain in the volume control section (the line stage if it was an independent preamp).

HK 990 Power Amplifier

The power amplifier is full complementary from input to output. There are eighteen transistors just to implement the voltage gain section excluding the class AB bias stage. This complexity ensures the open-loop distortion is low so only a small, constant amount of feedback, as dictated by Otala, is required around the complete amplifier from 20Hz to 20 kHz.

The differential pair at the front of the amplifier is biased by current sources and a buffer stage isolates the first gain stage from the second. The differential pair and second gain stages are both cascaded. Both voltage gain stages have local feedback (emitter degeneration). The four circuit techniques linearize the open-loop distortion of the amplifier and keep the return-loop gain to Otala’s desired minimum level. Just for reference, some AVRs over $1000 do a complete power amp with eight transistors including the AB bias and current gain stages.

Otala and other researchers showed much more current is required to drive a speaker than a resistor. In the HK 990, the current gain block is a triple Darlington terminating in five paralleled 15 amp continuous ON Semiconductor power transistors (MJL3281A and MJL1302A complementary pair) connected from one supply rail and the speaker terminals. These devices have 260V breakdown voltages.

Counting output transistors at the output of an amplifier is a futile exercise. You need to refer to the data sheet to find the short term and steady-state current each transistor can source or sink safely. The frequency at which the current gain goes to 1 (the point at which we could replace the transistor with a wire) should be near 30MHz. Current gain in the audio band should be greater than 20 when sourcing the maximum steady state current. Open-loop distortion under full load depends on the process technology. Typical specs sheets for an output transistor run more than four pages. Paralleling output devices has the disadvantage that the load the pre-driver sees is more difficult to deal with.

The HK 990 is dual mono down to its two transformers, a feature found on many older Citation products. All the metal in the transformers and output-stage heat sink literally weigh down this forty-five pound unit.

The HK 990 power amp uses separate transformer winding for the voltage and current gain stages. +/- 80V for the voltage stages and +/-60V for the output stages, as shown in the diagram above. This approach prevents the power supplies for the voltage stage from being modulated as the current stages send significant current to the speaker. A higher supply voltage for the voltage gain stage is rarely attempted because under a fault condition (shorted speaker terminals, for example) the number of pathways through which the transistors can be damaged multiplies. In a traditional amplifier with a high return-loop gain, the amplifier is relatively insensitive to modulation of the power supply (power supply rejection ratio). With the lower return-loop gain required by Otala, the modulation results in distortion.

Since spec sheets make a big deal of the size of the capacitor on the unregulated rails, I can report each cap on the +/-60V supply is 13600uF, slightly less than reported on the HK spec sheet. The size of these capacitors is important at low frequencies (20Hz) where the power supply is required to source or sink significant current in one direction as the sign wave remains at its maximum or minimum voltage level for a long period relative to the power supply refresh rate (120Hz). This is why THD goes up at low frequencies in some amplifiers when driving low impedance loads at the maximum output voltage swing.

The +/- 80V rails have 1000uF capacitors on the unregulated rails. This capacitor can be much smaller since the current drain on the 80V supplies from the voltage gain stages is smaller than what is flowing on the +/-60V rails driving the speaker.

The HK 990 spec sheet gives no FTC power rating into 4 ohms. The back panel of the HK 990 has a 1000 Watt rating for total power consumption from the AC line. Obviously, some of the power heats the amplifier owing to the efficiency of a class AB amp, but the construction points to the unit’s capability of providing almost 300 average Watts per channel (RMS power does not exist except in the mind of some marketing people. The engineers should get the blame for not proof reading the material. HK avoids this error.) into 4 ohms under FTC test conditions across the full frequency band. Why Harman does not supply such a key spec is a mystery. The mystery deepens because fans are mounted to the large heat sinks to aid in passing the FTC preconditioning test.

Harman Kardon was the first company to adopt the Audio Graph Power Cube measurement system. The unit tests for stability into inductance and capacitive loads (+/-30 degrees, +/-60 degrees) when the amplifier is driven to full power into loads as low as 1 ohm (magnitude of the load impedance). The test signal is 20 periods at 1kHz, with the power reported at a THD of 1% I do not have results for the HK990.

The Audio Critic had access to a Power Cube measurement system. Harman products always did well while many other higher-priced units failed. That said, some amplifiers with traditional circuit topologies also did well. Class D amplifiers tend to create poor Power Cubes.

The idle current of the output stage is very high. The unit gets hot with no input signal, so it is advisable not to rest another component on the HK 990. The high idle current is an attempt to reduce crossover distortion with the low return-loop gain specified by Otala. Doug Self has shown crossover distortion can be made extremely small at low idle currents.

Not a single bypass or blocking capacitor is found in the power amp. A DC Servo circuit is designed to prevent a DC offset voltage from appearing at the speaker terminals. A DC servo provides a compensating voltage at the power amplifier’s input to remove the DC offset at the output. The servo circuit is designed to only respond to subsonic frequencies and DC. The use of a DC servo to eliminate capacitors is not a universally accepted method to improve sound quality. Some argue the additional active circuitry for the servo is more audible than a well-chosen capacitor.

Be careful with the power amplifier inputs. If the input has significant DC, the amplifier will go into protection (note the DC over-current sensor in the figure. The DC over-current detector also activates if the speaker terminals are shorted or an internal component fails in the amplifier.). An amplifier with a DC blocking cap at the input would remove the DC.

New to the HK 990 is the ThermalTrak class AB bias stage. Normally the bias circuit has a diode placed on the heat sinks so temperature of the output devices can be sensed. ON Semiconductors place a diode on the same die as the output device (NJL3281D and NJL1302D) to improve thermal tracking (please refer to the figure above). The thermal time constant associated with the heat sink is eliminated. ON semiconductor has shown ThermalTrack reduces distortion.

Harman has a odd specification in their literature: 200 amp instantaneous current. At first glance, it appears to be a typo. Just considering the primary capacitor size and power supply rail, the capacitor would be discharged in 4msec by my calculations. This spec turns out to be an old measurement from the Audio Graph Power Cube test system mentioned above. The amplifier is connected to an 0.1 ohm resistive load and driven with a 10kHz (100usec) square wave. Audio Graph looks to have deleted the made in the current system. I am hard pressed to see the value of the specification now, but some readers with long memories are going to post me that The Audio Critic did report it when I was Technical Editor.

HK 990 Phono Stage

In his first preamp for Harman (Citation XXP), Otala addressed the challenge of achieving a flat return-loop gain across the frequency band. In a typical amplifier with a flat closed-loop response, this is readily achieved in a discrete circuit at the cost of distortion (attempts are made in the discrete opamp design to reduce open-loop distortion when the return-loop gain is purposely set to a low value as we saw in the power amp).

The problem is more difficult in a phono preamp with a varying frequency response (it follows the RIAA curve). The HK 990 uses a topology dating to Otala’s tenure. Twelve transistors form a transconductance amplifier. The transconductance amplifier differs from an opamp since it has a high output impedance. Think of the transconductance amplifier as an opamp without the common emitter output stage. It is called a transconductance amplifier because a change in voltage at the input (open loop) results in a change of current at the output.

Iout=VinG where G is conductance which is the reciprocal of resistance.

The RIAA network is connected around the transconductance stage. As the frequency increases, the passive RIAA network loads the transconductance amplifier and lowers its open loop voltage gain (the current output of the transconductance amplifier flows in the RIAA network giving rise to a voltage). Since the transfer function of the passive RIAA network is increasing with frequency, the total return loop remains constant at the desired level chosen by the designer.

The plot below was taken from the original literature for the Citation XXP preamp.

Low frequencies pose a problem with this approach. Here, a closed-loop voltage gain of 60dB is required to match the inverse RIAA curve for a moving magnet cartridge. The open-loop gain of the transconductance circuit loaded by the RIAA network should be 80dB at a minimum. This may not be achievable because the intrinsic output resistance of a real transconductance amplifier limits the gain. The result can be a drop in the gain at low frequencies and increase an distortion.

I do not have measurements of the HK 990 to confirm if it exhibits this problem. Some earlier HK phono stages hinted at the problem, but the transconductance amplifier on the HK 990 is significantly enhanced. It remains fully complementary. Current sources replace a resistor to bias the differential input stage. A buffer between the first and second gain stages prevents the second stage from loading the first. The first voltage gain stage design trades off degraded noise performance for improved open-loop distortion.

The DC servo used in the phono stage eliminates all bypass and coupling capacitors. Surprisingly, no DC blocking capacitor is at the input to the phono stage. Since a bipolar transistor is used at the input, a small current always flows in the cartridge. I have rarely seen this direct connection implemented unless the input transistor is a FET with an extremely low gate-current flow.

An open-loop emitter-follower buffers the transconductance amplifier from the load presented by the line stage. Almost all phono stages are in the non-inverting feedback configuration. Such a configuration is limited to a gain no lower than one. The RIAA curve should continue rolling off and not stop at unity gain. A passive low-pass filter in this stage corrects the problem (The Audio Critic, Issue 18, page 16). This and other phono stage design issues are addressed in Chapter 7 of Small Signal Audio Design by Douglas Self (Focal Press, 2010).

A 330pf capacitor is at the input. This high value exceeds the specifications of most moving magnet cartridges when the turntable wiring is included. At this price, I would like to see a switch on the back to offer different loading options. In the absence of that, a low-valued capacitor is advisable since the value can be raised externally. In contrast, nothing can compensate for too much capacitance at the phono input.

Moving coil cartridges see a pre-preamplifier that is a two-stage open-loop design (no feedback). Since the signals from the moving coil cartridge are so small, the transistors almost act as linear devices; hence, no feedback is required to linearize the moving coil stage. Other designers might have employed feedback to reduce the distortion to lower levels. The pre-preamp has no DC servo. A standard DC blocking capacitor is at its output. In the moving magnet stage no blocking capacitor is at the stages input. Again its absence causes a small DC current flow in the cartridge.

HK 990 Headphone Stage

The high performance Texas Instrument TP6120 obviates DC coupling or blocking capacitors in this stage. The TP6120 has very low distortion and noise. It has a current feedback topology and 1300V/us slew rate. I have no idea why it is designed with a slew rate 100 times larger than what is required. Without circuit details, I cannot conclude if Otala would approve, but with a THD under full load approaching 0.0002% at 1 kHz, I doubt this design has low return-loop gain.

One issue I identified involves the calibration microphone connected using the headphone output. If nothing is done, the microphone could potentially be damaged when plugged in while a large level signal is at the headphone jack. The headphone amps are disconnected when the HK 990 is put in calibration mode, but it is unclear if they are disconnected automatically when the microphone is inserted and the unit is not set to calibrate. I suspect all HK units with room calibration have the same issue but they use a small IC to drive high efficiency headphones. The TP6120 can deliver much more current and only a 20 ohm resistor is between it output and the headphone jack. The instruction book does not caution about these circumstances.

HK 990 Analog Circuitry Connected to the DACs

The analog circuitry in this stage is a typical design and does not reflect any influence of Otala so I will not spend much time discussing it.

The Analog Devices AD1955 has a current source output. No on chip opamps are present.

Analog Devices OP275 and TI OPA2134 are used. The characteristics of these opamps are more conventional than those of the AD825. I suspect the higher power supply rejection, lower noise and lower input capacitance of these parts may have played a role in their selection. The absence of discrete circuitry is going to disappoint purists wedded to Otala design.

A DC servo circuit senses the DC offset voltage at the output of the DAC signal chain and introduces an offset current at the DAC to correct this (recall the DAC has a current output). The topology eliminates all coupling and DC blocking capacitors from the signal chain This is the first time I have seen a DC servo integrated in the digital-to-analog conversion block.

The output of the analog electronics connected to the DAC goes directly to the volume control bypassing the buffer in the direct path. In total three active opamp stages are present before the signal enters the power amp. One DC blocking capacitor is in the signal path. From the perspective of an engineer who does not slavishly adhere to Otala design rules, the execution of the direct analog and DAC paths is extremely well done.

Conclusions About the HK 990 Circuit Design

I wrote much of this material in October 2010. At the time, my sole reference was the service manual (the actual unit arrived several months later). I intended the piece to be a preview of the HK 990, which had just been shown at CEDIA. Since then, the HK 990 received a positive review from Tyler Stripko and at least one other professional reviewer. It is rare for a relatively low-priced unit to achieve such consistently positive reviews. I will not comment about my impressions of the sound of this amplifier, other than to note that Tyler’s positive sentiment may be justified by the work of Otala. Every design decision is backed up with some quantitative analysis. The HK990 is a polar opposite of designs offered by others where the design techniques are more closely allied to black magic than solid engineering and expensive metalwork wraps the questionable design to give the look of ultra high tech. A five-figure price is tacked on to enhance the aura of ultimate performance. Poor measured results are often a consequence when black magic replaces science. The component may sound different, but different is not necessarily better.

HK 990 Dual-Domain Tape Recorder Outputs and Tape Monitor Details

Having been involved with tape recorders since reel-to-reel was the state of the art. I invested considerable effort in testing the dual-domain tape recorder interface function. Conceptually, a dual-domain tape output path should be a must-have for those seriously rooted in recording. With a dual-domain configuration, both analog and digital inputs appear at the analog tape output jack (to recorder in) for the two tape recorders (Harman calls these CD-R and Tape). Analog and digital inputs also appear on the single record digital output (to digital-in on the CD-R or MiniDisc player). The rear panel section with the tape outputs is shown below.

Harman allows the tape outputs to record a source (analog or digital) independently of what is heard on the speaker. This feature is relatively common in analog integrated amplifiers and preamplifiers replacing the single tape monitor button.

Listening to one input and taping another provides a tape monitor function. For example, one might tape the Tuner with the cassette deck connected to Tape using this sequence of operations:

  1. Select Tuner using the record-out selector.
  2. Select Tuner on the remote control to listen to the source through the speakers.
  3. Press Tape on the remote to hear the recording off the cassette deck.

Even with a CD-R, monitoring is important. I cannot tell you how many times I thought I hit the record button, but inadvertently hit Play instead. With the meters moving, everything looks OK, but it’s not. Another example where monitoring could save the day is when the Tuner is selected as record-out and not Phono that you wanted to record; again, the meters are moving but you are recording the wrong thing.

Source selection for the tape recorders is only available on the front panel, making the selection more than a nuisance when all the buttons on the front panel are similarly-sized with low-contrast lettering.

To select a recorder, the Record Output button is first selected, then the two Source Select buttons are toggled, and finally Record Output is re-pressed to return the source selector to its normal function so the speakers are operative. From the remote, toggling the source select for the output to the speakers is unnecessary. Each input is assigned a button.

Closely-spaced buttons may cause an adjacent button to be depressed. Put your finger a little to the right of source select and you press the input assignment setup button. Cancelling that is no fun.

Many AVRs offer one or two tape outputs (with composite video outputs also supplied for at least one). These outputs are analog and transmit only analog inputs. No monitor function is available.

You can try to mimic the Harman’s functionality using the Room 2 analog outputs on the AVR, but care is required as there is no protection for self loops under this setup: say goodbye to your tweeters were both Tape for Room 2 and source selected at the same time with the recorder activated. The oscillations are typically a square wave at the maximum swing of the tape recorder.

You may be able to work around this using the AVRs Room 2 input assignment GUI. One might try to assign Room 2 Tape In to an unused analog input. I make no guarantees you can perform a setup with an AVR Room 2 that will insure a disaster will not occur. I play it safe and live without the monitor function with an AVR. I use headphones connected to the recorder to do the monitoring function.

Almost all AVRs provide only analog inputs (single domain) at the Room 2 outputs.

Tape I/O signal flows at the block diagram level

This diagram is similar to the block diagram of the digital input selector presented in Part 2, but the circuitry to support tape recorders has been included. The yellow box highlights the added circuits. The Texas Instrument SRC4392 multi-function chip has two digital input selectors. As can be seen one selector is used for the tape output path and one for the main path. In the tape output path the selector only routes SPDIF signals (green) and does not recover the PCM data as it does for the main path. A second set of input selectors are required for the tape recorder output path because the digital input selected is different from what is being sent to the speakers.

An extra DAC is required in a dual-domain tape system to convert the digital inputs to analog. This is highlighted in the yellow box. The DAC for the analog record outputs is part of a multifunction AKM 4683. The performance of the AKM DAC is lackluster, with a worst-case dynamic range equivalent to 15.5bits and distortion of 13 equivalent bits. The AKM 4683 also houses the requisite SPDIF receiver.

Older cassette and reel-to-reel machines need the analog output. The quality of the conversion of the digital inputs to analog is not that important since all CD-Rs and MiniDisc recorders have digital inputs; however, a problem with this assumption in the case of the HK 990 will be identified below.

A designate the diagram above BLOCK DG.

The analog selector block is shown in this diagram, with the circuitry to support tape recorder in yellow. Like the digital path (BLOCK DG), the analog input selector has second set of switch selectors that aggregates the analog inputs for the tape recorder outputs. Everything in this tape output channel (in the yellow box) is only AVR grade: the switches, for example, are MOSFETS, not relays.

The selected analog input must be converted to digital to provide a digital tape output. The ADC should be of high quality because it replaces the ADC in the digital tape recorder. Unfortunately, the HK 990 disappoints. The ADC is also in the AK4683 (with an ADC and DAC, it is called a CODEC). The ADC has a worst-case dynamic range equivalent to 15.5 bits and distortion of 13 equivalent bits.

You may ask why the output of the high-quality Cirrus CS5361 ADC in the main path (in the upper right of the diagram) is not used instead of adding an ADC just for the tape recorder outputs. The principle justification is the bifurcation of the input to be recorded from what is played on the speakers.

A designate the diagram above BLOCK AN.

This block diagram illustrates the conclusion of the recorder output’s journey as it makes its way to the rear panel RCA jacks. This diagram clarifies the dual-domain aspect of the tape recorder path. Both digital and analog outputs are shown at the right. Depending on the component that the user has selected to record, a signal from the analog block (AB) or digital block (DG) is sent to the output (signals entering at right).

Note the switch prior to the two analog outputs CDR-Out and Tape-Out. The switch prevents self-oscillation. When listening and recording from CDR, the switch mutes CDR-Out, preventing a self-oscillation through the CDR recorder. Tape-Out is muted under similar conditions.

The digital Coaxial Out lacks the switch. The single digital output is live when you select to record from the CD-R or the Tape input. In contrast, the analog output for CD-R (Tape) mutes when to-record CD-R (Tape) is selected. Mixing analog and digital connections in a recorder can cause a self-loop and high-level oscillation that may damage your speakers.

Proper Connection of the HK 990 Tape I/Os to a Digital Tape Recorder to Avoid Self-Oscillation

When connecting the HK990 to a tape recorder, I advise one of two methods to prevent self- oscillation:

1) Use only analog cables for the input and output connections.

Given the quality of the ADC in the HK 990 (AK4683) little is lost since the ADC in your CDR is likely to be as good as the one in the HK 990, especially if the ADR a semi-professional unit. Coming out of the CD-R analog to the HK 990 does degrade the signal since it travels through the DAC in the CD-R and the ADC (AK4683) in the HK 990. This is a redundant DAC – ADC path.

2) Use only the digital cables for the input and output connections.

Self-oscillation is avoided by coming into the HK 990 from a recorder’s digital output (in the all-digital hookup) because both digital inputs mute when record selector is set to CD-R or Tape. In this manner, the oscillation is broken on the input connection for the digital inputs and the output connection (discussed above) for the analog connection. This is what makes mixing analog and digital connection so dangerous.

By simultaneously muting both CD-R and Tape digital inputs, tape-to-tape copying from CD-R digital in) to Tape is enabled. This is probably related to the Digital Millennium Copyright Act, whereby Tape-to-Tape copying is allowed with only analog interconnections.

At first glance, an all-digital connection of the CD-R appears preferable. The all-digital path eliminates the DAC – ADC redundancy on the recorder output side. However, conversion quality is crimped because the AK4683 ADC replaces the ADC in your CD-R.

With only one digital output, I do not understand why it is not marked digital output for CD-R and made to mute when CD-R is selected as the analog CD-R does. This should be a straight-forward software fix to allow mixed analog and digital connections.

The digital output is limited to a 16 bit depth and 48kHz sampling rate. Consumer digital recorders would mute if high resolution SPDIF data was allowed to appear at the output.

Conclusions About HK990 Tape Recorder Functionality

Harman is to be congratulated for providing the dual-domain tape recorder interface. Regrettably, the front panel controls make it difficult to use. The quality of the internal ADC and DAC in the tape recorder path is disappointing. Mixing analog and digital connections to a CD-R or MiniDisc recorder can result in potentially damaging self-oscillation.

Overall Conclusions About the HK 990

For those who started with Tyler’s review and worked their way through this three-parter, 20,000 words have passed your eyes. Only a very special product requires that level of analysis. There is no comparable for the HK990. One could say that an equivalent could be crafted with three or more boxes. Unfortunately, this is not a viable option because many different functional blocks inside the HK990 interface with each other in ways that cannot be replicated with RCA cables running between multiple external boxes.

As with any debut product, the HK990 has some glitches. The front panel controls are very difficult to manipulate and the room-correction system has some software bugs. Putting these issues aside, the HK990 is a revolutionary product that will be on the list of the 100 most important audio components ten years from now.


Video EQ
Written by Ron , August 09, 2011

Ideally, and as described in the article, it is always best to calibrate the display first with an accurate external pattern device, however, I have found there IS an option(for an extra $1200) for multiple sources with the use of the AV Foundry Video EQ Pro. If required, this allows an individual to calibrate Grayscale, Gamma, Luminance AND Color Management for up to FOUR individual source components which would go a long way toward taming the differences with any potential colorspace issues.


Written by Jerry , August 12, 2011

How about revealing what player A and B are? It might be interesting.


Where is the test for the PS3?
Written by Shawn , August 13, 2011

Where is the test for the PS3?



    Players and PS3
    Written by Ron Jones , August 29, 2011

    As for what players A and B were, I would assume these were the Oppo and Sony players are tested in Part 1 of this story. As shown in Part 1 the Sony standalone players appear to be doing some sort of mapping betweeen the grey levels recorded on the disc and what is actually output via HDMI. When I first read part 1 a couple of months ago a bell immediately went off for me as it explained why I had seen such non-linear gamma results (with a curve that looked like an inverted ‘U’) when calibrating a JVC projector and using a Sony BDP-S470 player playing the HD Calibration disc as the source. I went back and reviewed results for a couple of other displays where I had used PS3s (both older Fat models) playing a HD calibration disc as the signal source and these did not show the poor gamma curve. So I don’t think the PS3 has this same issue as the Sony standalone BD players. However, you probably should not set the PS3 to use ‘RGB Full’ as that mode for the PS3’s HDMI output is mapping grey scale levels between the video standard (16-235) and the PC standard (0-255). I have my PS3’s set to output in Y Cb/Pb Cr/Pr mode.


    Written by ChrisHeinonen , August 30, 2011

    While we are still working on the PS3 data, you certainly should use RGB Full if using RGB, and SuperWhite if using YCbCr. Otherwise values that should be 0 are mapped up to 16, and values that should be over 235/240 are truncated at those points. Some people will tell you that’s fine since there should be no video data in there anyway, but if you calibrate correctly you will have that data hidden by the Brightness and Contrast of the display, but you will still get the correct data from the player. RGB Full did not look to do any remapping in our testing unless RGB Full wasn’t enabled.


    PS3 Correction
    Written by ChrisHeinonen , August 30, 2011

    Sorry about that last post, I shouldn’t comment before coffee. Superwhite on the PS3 is good, RGB Full is bad would be a good general view as RGB Full does remap 16-235 to 0-255. If you’re using your PS3 on a computer monitor that’s calibrated for that range you’d want to enable this, but if you are using it on a video display calibrated for the normal video range, it would not be good.


    PS3 Correction
    Written by Hydrosaure , September 02, 2011

    I agree with you Chris.

    The PS3 does a terrible job at converting YCbCr to RGB Full range.

    Even for games, you’re better off staying in Limited range and let you TV/scaler do their job.


    More Please
    Written by Donnie , March 22, 2012

    Okay , now you’ve whetted our appetites, can we get some more “correct” players and shootout data?. We’re waiting!!



    So HDMI is not simply Video Pass through?
    Written by Percy Mistry , April 05, 2012

    I was of the impression that using HDMI as a transfer medium allows the raw data from the disc to be sent from the source (player) to the destination (TV) without any processing. Kinda like a pass through – the player is nothing but a true “transport”. So in case of bluray it would the pure 1080i data going to the TV. Apparently that doesn’t seem to be the case! The players are messing around with the data. But WHY ? Is it not possible to send the data without modifying it over HDMI to the TV?


    Video Pass Through
    Written by ChrisHeinonen , April 05, 2012

    All video eventually winds up as RGB data, since that is what a display presents. Data is all stored on Blu-ray discs at YCbCr 4:2:0 in order to save space, as storing it as raw RGB (4:4:4) would take up far more space, and Blu-ray is already space limited. Also, all consumer devices work with 4:2:2 or 4;4:4, so it has to be converted before it is sent over HDMI.

    Most Blu-ray content, at least for commercial film, is stored as 1080p24, so it is natively progressive. TV shows and concert videos are the most common 1080i60 sources, but they too are 4:2:0 encoded for Blu-ray.

    It really isn’t possible to send 4:2:0 to a commercial TV. You can find some displays in the commercial realm that might handle 4:2:0, but they’d also be likely to have an SDI input on them. You can slso find modified Blu-ray players that can do 4:2:0 over SDI, but you’d need a processor (such as the Lumagen Radiance line) with an SDI input to handle that 4:2:0 as your display can’t. At that point you’re still getting a conversion to 4:2:2 or 4:4:4 or RGB, which is what we’re testing for anyway.

    So the simple answer is, no. Until we have a source media that uses something like 4:4:4 or 4:2:2 for the source content (perhaps 4K can do this, and have a larger color space like DCI) it will probably remain the case as well.