by Toni Kemhagen
I thought that my portable headphone amplifier with crossfeed would be an interesting project for DIYers. My intention when I started the project was to build a “state of the art” amplifier, with sound quality being the first priority, but still be portable and have relatively long battery life. I am very pleased with the result. The only disadvantage is that the amp is a little larger and heavier than would be optimal for a portable due to the fact that it uses four 9V batteries. The completed amplifier is still very small – it measures just 80mm x 150mm (3.2″ x 6″).
The headphone amp is a 2-stage design with Analog Devices opamps. The are two completely separate ±9V battery supplies, one for the input buffers and voltage amplifiers and the other for the output stage only. I added a crossfeed filter designed by a Swedish engineer named Ingvar Ohman. The level of crossfeed in the Ohman filter is higher, for example, than in Jan Meier’s crossfeed filter, but has better effect.
The Headphone Amplifier Circuit
Different opamps are more or less sensitive to impedance mismatching. This headphone amp design corrects for opamp input errors and distortion due to impedance mismatching, so that both the + and – inputs of the AD823 opamps see the same impedance. For example, in the “System Applications Guide” from Analog Devices, page 8-71, there is a distortion test with the OP275 opamp that shows how a mismatch of 10K ohms/910 ohms at the inputs can triple the device’s output THD+N. Also see Walt Jung’s article, Minimizing Input Errors, in Electronics Design, December 18, 1998. [Editor: And see, Designing an Opamp Headphone Amplifier.]
The headphone amplifier has 2 stages. The Analog Devices AD823 opamp in the first stage works as an input buffer with a gain of 2. The 2.49K resistor compensates for the feedback impedance. The compensation network (0.1uF + 1.87K + 100K) at the – input of the opamp balances the impedance of the input network (with the pot rotation set at the usual position for listening).
The input compensation network is a compromise. With the 0.1uF capacitor, there is still a slight mismatch in the 30Hz to 500Hz range. The optimal would be 1.0uF which would match the impedance over the entire frequency range. I use 0.1uF because there is limited space in the enclosure and because it is most important that the impedances are matched at DC and upper frequencies. Otherwise, if the enclosure had enough space, I would use a 1.0uF capacitor. Also, the impedances are matched with the crossfeed on. When the crossfeed is off, there is a slight mismatch, but I always listen with the crossfeed on.
I do not know how much the AD823 is sensitive to impedance missmatching. I have not tried using the AD823 without the compensation, only with it. In my project, I have taken every practical step to minimize distortion.
The second and third stage (AD823 and AD812 opamps) is a composite amplifier with a gain of five. The inputs of the AD823 in the composite amp are impedance matched to compensate for the feedback network and the crossfeed network. The local feedback at the AD823 in the second stage sets the open-loop bandwidth to about 100kHz. The advantages of a composite amplifier are:
- No thermal coupling from the output stage to the controling input amp.
- The possibility of having separate power supplies to the AD823 and the AD812 opamps, so that the AD823s can get the cleanest power. Also, I think this amplifer is much less sensitive to the high impedance of the batteries.
- The ability to combine the JFET-input of the AD823 and the current feedback operation of the AD812.
The headphone amplifier has a total gain of 10. If you want to modify the total gain to 5, for example, the best way is to change the gain of the first stage to 1.
The Ohman Crossfeed Filter
This crossfeed filter that I use is a design by a Swedish audio engineer named Ingvar Ohman. The design was published in an article called “Den Lilla Stereo-kontrollboxen SP12” in the December 1994 issue of the magazine “Musik och Ljudteknik” (“Music and Audio Technical Society”). The “SP12” in the title is nothing particular – only numbers and letters to give the project a name. In the article, Ohman is tried to be a little funny and explained that SP12 could stand for “Stereo Processor According to a 12-Year-Old Idea.”
The graph above shows the frequency response of the Ohman filter with the right channel as the main signal and the left channel as the crossfeed. The filter has an input impedance of 210K ohms to 16K ohms from DC to beyond the audio range. The output impedance varies from about 50K ohms (DC) to 3.5K ohms (high frequencies). Because the filter has a high output impedance and because the input and output impedances change so much with frequency, the filter should be isolated between two buffers. There is no voltage drop – a mono signal goes straight through without loss.
The maximum delay between the main signal and crossfeed is about 250uS at frequencies less than 100Hz. Ohman said that it is more important to have a correct frequency response for the crossfeed, and less important to have time delay. The 250uS that separate the right and left channels after crossfeeding correspond to 8.5cm of travel in air. For a person with ears that are 17cm apart (the average), the time delay exactly corresponds to a listening angle of 30 degrees for loudspeakers (which is the stereo standard).
Though the crossfeeding circuit may seem to make low frequencies almost mono, it is time-coding at low frequencies that is totally optimal and what is heard in reality. Low frequencies arrive at both ears with practically identical volume, not only the in-front sound, but also sound incoming at 30 degrees, which corresponds to about 250uS delay.
Ohman tried to add time delay to the higher frequencies, but did not like the sound (similar to a commercial crossfeed product that had full-spectrum time delay) due to the serious comb filter effects on mono signals. In order to compensate for too little time delay, extra channel separation in mid frequences is added – a trick to fool the human ears to believe that the time delay exists.
I think Ohman developed the filter for the recording engineers that have to use headphones. It matches most people’s hearing. He made many calculations and measurements and tests on many people. It has a larger amount of crossfeed than, for example, the filter by Jan Meier. Anyone who finds the Meier filter to be too subtle should try this circuit. The result is alot more depth in the audio image! I find that the Ohman crossfeed circuit is especially good with classical music, because it gives more depth to the audio image. Even if you do not build the portable headphone amplifier, please try this crossfeed circuit.
The Power Supply
The output current buffers have their own ±9VDC power supplies, total separate from the ±9VDC power supplies of the voltage amplifiers and input buffers. The separate supplies mean that the input buffers and the voltage amplifiers that control the output signal have a cleaner power source. I use rechargeable NiMH 9V batteries (150maH capacity). You can use alkaline batteries and get longer battery life, but they will be rather expensive of course.
The AD823 has a low quiescent current of 2.6mA and a minimum power supply of ±1.5V. The AD812 idles at 4mA, has a short circuit current of 100mA and a minimum supply voltage of ±1.2V. There are a total of four AD823 opamps, so the battery supply for these devices will last about 150mAH / (4 * 3mA) = 12.5 hours. The battery life for the two AD812 opamps will be 150mAH / (2 * 4mA) = 18.75 hours in the idle state. I use a 4PDT switch to turn the power off – one pole for each battery.
The AD812 power supply will drain faster when driving headphones, depending on the headphone efficiency, headphone impedance and the listening volume. I use a Sennheiser HD600 (300 ohms, 97db/mW) and the batteries go out at approximately the same time. But with 30ohms cans, maybe the batteries that go to AD812 drain faster. A 30-ohm headphone will drain ten times more current than a 300-ohm headphone (with the same sensitivity).
I have checked the battery life more exactly. This data is for normal listening volume. The data is calculated from current measurements:
|AD812 batteries||AD823 batteries|
|300 ohms (HD600)||18 hours||12 hours|
|60 ohms (HD570)||16 hours||12 hours|
|30 ohms||14 hours||12 hours|
In general, the batteries to AD823s drain a little faster than batteries to the AD812s, but the battery life for the AD812 batteries depends on the volume level and type of music being played:
|Classical (low volume)||Rock (high volume)|
|300 ohms (HD600)||19 hours||16 hours|
|60 ohms (HD570)||18.5 hours||10 hours approx.|
|30 ohms||18 hours||7 hours approx.|
I am using a “wall-charger” to charge the NiMH 9V batteries, which takes about 12 hours. The schematic above shows a recharger circuit for the headphone amplifier that equalizes the battery life when the amp is turned off and will fit in the amp’s enclosure. (I have not built this circuit yet.) Most people do not listen as long as 12 hours at a time. On average, someone may listen for a few hours and then turn the amp off. During the time between the listenings, this circuit will equalize battery levels, which will be ready for the next listening session with the same charge status.
For example, if the batteries to the AD823s have drained faster, then the AD812 batteries will charge the AD823 batteries through the 2 x 68 ohm resistors. And after a few hours, the charge status for all of the batteries will be equal. If all of the 9V batteries are at the same charge, it is simple to recharge all batteries at the same time using the AC charger. If you recharge with low current (aproximately 15mA), the charging time will be about 14 hours.
I mostly get my parts from a distributor here in Sweden: ELFA. They have most of things DIYers need, and you do not have to be a company to buy from them. But they do not have the AD823 and AD812 opamps, and the output inductors. The capacitors in the signal path are Icel 1% polypropylenes. The power supply capacitors are Panasonic FC Electrolytic and Wima polyester MKS2. The resistors are all 0.6W, but I think 1/4W resistors will work too. The volume control is a Bourns model 91, plastic type, 10K log taper (ELFA order number 64-256-07) and costs 142,5 SKR (approx. $15 US). The 4PDT switch for the power supply (ELFA order number 35-203-01) costs 99,25 SKR (approx. $11 US).
The output inductor can be the Miller 4608. I wound the 3.9mH inductor myself, because at the time, I could not find anywhere to buy one. Here is how to wind the inductor:
- Put a little grease on a 5.5mm drill bit.
- Shrink a small length of shrink tubing on the drill bit (at least 17mm when shrunk).
- Wind 31 turns of 0.4mm diameter, enamelled copper wire around the tubing – tightly and the every turn tightly together. Keep the length of the coil to 14.5mm.
- When done, hold both wire ends in left hand and apply a few drips super glue along the length of the coil. Hold it a few seconds until the glue gets hard.
- Slide the shrink tubing off the drill bit and trim to about 16.5mm.
The final specs of the inductor are:
- inner coil diameter: 6.7mm
- outside coil diameter: 7.5mm
- coil length: 14.5mm
- 31 turns of 0.4mm enamelled wire
The headphone amp has a separate ground plane, because it is easy to ground all components. On the downside, it is not so easy to modify the circuit, because you must desolder all of the ground connections before the ground plane can be removed. The enclosure is grounded at the backplate via a 10ohm resistor connected to the ground card. The 10 ohm resistor isolates any RF interference that the enclosure picks up from the air.
The size of the circuit card is 76.5mm x 100mm. The size of the ground card is 75mm x 100mm. The ground plane is a standard copper pc board. After that I decided where all the holes should be, I held the main pc board and the ground plane together and drilled through both boards at the same time. Then I separated the boards and soldered all the parts to the main board first and then to the ground plane.
If a component has leads that are too short, you can solder a longer lead to the component to be able to solder it to the ground plane. It is hard to solder onto the large area of copper foil on the ground plane, because it dissipates the heat of the soldering iron quickly. It is good practice to mill a bit around each solder point to reduce the heat dissipation.
I do NOT recommend the ground plane card for the beginning DIYer. The better way to have a ground plane may be with a double-sided pc card. If I build another amp, I will not use a separate ground card. In this case, I will try to make an ordinary star ground on the card with the components.
The enclosure is the Bobla “Alubos” model ABPH 800-0150 and measures 80mm x 150mm x 32mm (3.2″ x 6″ x 1.28″). The box and the front and back plates are sold separately at ELFA:
- Box: ELFA order number 50-792-49. Cost: 212,5 SKR (approx. $23 US)
- Front and back plates: ELFA order number 50-793-06. Cost: 73SKR (approx. $8) each.
The total cost is $39 ($23 + $8 + $8) – rather expensive but very nice! I find this case beautiful, and the four 9V batteries fit perfectly. It has a rubber rings around the front and backplates, so the case will not scratch the table.
The finished amplifier weighs 500g mostly due to the batteries. I have only used the amp with my home stereo and do not need a portable amp at this moment. When it is used as a portable amp, I do not think it is too heavy to carry, if the listener is interested in high quality audio.
I have tried the amp with the Sennheiser HD600 (300 ohms) and the HD570 (60 ohms), and the amp works great with them. The sound quality is very good. I have not been able to find any distortion. You can hear a very small amount of noise, if you turn the volume up to maximum with no music playing. I do not think that is a problem.
The audio image of the Ohman filter has very good depth. It has more crossfeed than Jan Meier’s filter. With the Meier filter, I still have an “in-head” feeling, and I do not feel satisfied until I have enough crossfeed. But, with the Ohman filter, the ambience is reduced. On classical recordings, I get a feeling of depth in the audio image that outweighs the reduced ambience.
The Ohman filter works best when the recording has wide stereo imaging and lots of ambience. At first, when I listen to some recordings with very narrow stereo image, I think it sounds a little dull. But when I have been listening for a while and get the feeling for the better audio image and can hear the instruments in their more proper place, I prefer the Ohman crossfeed in the long run. In other words, it is nice to move the ambience that you hear in an extreme stereo image to its proper position. You then can get a feeling of the original ambience in the audio image. It is also more relaxing and comfortable to listen with Ohman crossfeed.
It works on pop/rock music also. But in general, in pop/rock music recordings, there is no natural audio image from the place where the recording took place, as there is in a classical recording, when you can hear the ambience from the room. So, if there is very little stereo separation in a pop/rock recording, I do not use the crossfeed. For me, the lack of ambience in the recording demands a little wider stereo image.
But with most pop/rock recordings that have a very wide stereo image – with channels separated in each ear, this crossfeed is perfect!
Well, those are my thoughts, but of course, it is a matter of personal taste.
c. 2001 Toni Kemhagen.
1/27/01: Gus Wanner has prepared an Microsoft Excel application for simulating the operation of the Ohman crossfeed filter, so that DIYers can test the effects of component value changes on the filter’s frequency response, time characteristics and channel separation. Wanner provides simulations for 3 levels of crossfeed (3dB, 6dB and 10dB). He writes:
The attached Excel 5.0/95 spreadsheet is an electrical analysis of Ingvar Ohman’s crossfeed network as published in Toni Kemhagen’s article on construction of the Lindesberg Portable Headphone Amplifier With Crossfeed. The spreadsheet includes plots of left and right channel output levels, channel separation, interchannel delay time, network input impedance and network output impedance. I have also made included entries for source resistance and load resistance.
This crossfeed network is interesting in that it is the first one I have seen which deals with the problem of excessive high frequency channel separation when listening to normal stereo recordings with headphones. Since the optimum value of interchannel crossfeed is different for different recordings, I have included additional network analysis sheets for 3dB, 6dB, and 10dB low frequency channel separation. The shape of the channel separation versus frequency characteristic for these alternative networks follows (as closely as available component values permit) that of the original Ohman network specified in the referenced article. Since the crossfeed resistor (R4) is the same for all networks, a simple 2 pole 5 position switch would provide the ability to switch between the different networks for different degrees of crossfeed.
I appreciate the need to implement source impedance matching for opamps which are sensitive to source impedance when used in the non-inverting mode. Obviously, if source impedance is not constant (as with a gain control potentiometer at the input of an amplifier or, as here, with multiple crossfeed networks), source impedance matching becomes a major problem. A better solution would be to use opamps which are not as sensitive to this phenomenon. The manufacturer’s data sheets generally provide this information as part of their application notes.
One note about these networks – as currently parameterized they are very sensitive to load resistance. It might be desirable to reparameterize the networks by dividing all resistances by 4 and multiplying all capacitance values by 4, thus lowering the output impedance by the same factor. If FET-input opamps are used, the existing values will not be a problem.
The worksheets are protected (without a password) to prevent accidental deletion of calculations. Input signal magnitudes, source and load resistances are entered on the “original” network page and are automatically carried over to the other pages. Note that these spreadsheets make use of the Excel complex number capabilities which are included in the “Analysis Package”. This comes with the Excel package but may need to be manually installed using the original Excel installation CD or floppies. Current editions of Quattro Pro also include complex number analysis capability and may be able to load this spreadsheet.
5/29/01: Gus has prepared a new version of his Ohman Filter Simulator. Computationally, it is the same as the earlier simulator, but includes a schematic for adding a variable Ohman crossfeed filter to the Landesberg (or any other amp). Thanks to David Richard Meddings for the submission.
10/21/2001: The author presents an update to his amplifier design incorporating a sound image width control that operates separately from the Ohman crossfeed filter. He does not recommend modifications to the Ohman filter itself as a means of changing the width of the sound image.
- It is wrong to change the setup of the Ohman filter when you want to tune the stereo image. The Ohman filter is made to mimic a normal loudspeaker setup, and the differences in people’s hearing normally deviate only few percentages from the Ohman filter response. Modifying the Ohman filter’s response is not an option because then the filter is not an Ohman filter anymore; it’s just an “effect-box” that does something strange that may sound good to the user. Only one setup, the original Ohman filter setup, is to be used as a “true headphone monitoring system.”
My variable Ohman filter uses a stereo expander in the feedback path of the input buffer before the crossfeed processor, so that you will be able to tune the stereo image without changing the setup of the Ohman filter. Imagine that you are listening with loudspeakers and you listening to a recording with a narrow stereo image. When you put the loudspeakers farther apart, the time difference between the ears then has to increase. You see the same thing in the graphs below.
The filter takes the signal from one channel and puts it at the other channel’s negative input and, therefore, amplifies the differences between the channels, like a spatial expander. When the switch is set to “original,” you have the original Ohman crossfeed network working. When the switch is set to +1,+2, +3 and +4, both the stereo expander and the crossfeed are working, and the width of the audio image increases more and more. The widest sound image is at setting +4. When the switch is set to “stereo”, both the stereo expander and the crossfeed are disconnected, and the amp is in normal stereo mode. The switch is a 6-position, 2-pole shorting type – for example the ELMA typ01 (Elfa order no. 35-493-18).
I am very pleased with the result of this Variable Ohman crossfeed filter. I find it most useful to fine-tune the stereo image when the image is too narrow. It works with all types of music. But the most recordings donīt need to expanding, only if the sound image is too narrow or “boring”.
Frequency Response of Variable Ohman Filter (Freq. vs. Volts)
Time Response of Variable Ohman Filter
Channel Separation of Variable Ohman Filter
Frequency Response of Variable Ohman Filter (Freq. vs. dBV)
11/17/2002: Corrected polarity of electrolytic caps for negative power supply.