New Aurora converter - help wanted with review

[tubesrule]

Hello all,
Sorry for the late entry into the thread, but I just got my third kidney stone :-O

Anyway both Jeff and David have correctly spoken. First, the composite output from the converter is a simple NPN emitter follower. Since the main goal of this design was to offer the best conversion quality with a built in modulator, very little cost was allocated for the composite output which would probably never be used. Since the RF modulator is built in, there is little need for the composite output. It was added at the last minute since there was physical room in the box for the connector. Also, as Jeff has pointed out, when driving a correctly terminated 75 ohm load, the output performs properly. It's main benefit is that it offers the full video output bandwidth as the sound trap filter is after this circuitry.

For the modulation specs, the depths quoted in the data sheet are for a 1Vp-p input, typically from a video source, and account for variations in the modulator IC's themselves. As David has stated, these modulators can be driven to full 100% and beyond which will eventually result in a reversal. I do operate the AM audio modulator in the negative video mode as David mentioned, and supply a biasing pot in the audio op-amp circuitry to set the nominal level.

Hope this helps.
Darryl

[oldeurope]

Originally Posted by tubesrule

Hello all,
Sorry for the late entry into the thread, but I just got my third kidney stone :-O

For the modulation specs, the depths quoted in the data sheet are for a 1Vp-p input, typically from a video source, and account for variations in the modulator IC's themselves. As David has stated, these modulators can be driven to full 100% and beyond which will eventually result in a reversal. I do operate the AM audio modulator in the negative video mode as David mentioned, and supply a biasing pot in the audio op-amp circuitry to set the nominal level.

Hope this helps.
Darryl

Thanks for answering here and best wishes to you, Darryl.

Ah, you are right.

I did not know that the peak white clip (PWC=1) can be switched off in the negativ vision mode (SYSL=0). This makes it possible. So the values in Table 21 "PAL video modulation depth" are not guilty for your sound application.

Well done!

Kind regards
Darius

[oldeurope]

Originally Posted by tubesrule

Hello all,
...very little cost was allocated for the composite output which would probably never be used. Since the RF modulator is built in, there is little need for the composite output. It was added at the last minute since there was physical room in the box for the connector. Also, as Jeff has pointed out, when driving a correctly terminated 75 ohm load, the output performs properly. It's main benefit is that it offers the full video output bandwidth as the sound trap filter is after this circuitry....Hope this helps.
Darryl

Good morning,
the vision output will be used to record something on for example vhs tapes or to drive another modulator. Often old vcrs are used and they sometimes have a cap and resistor terminated input. This design is used to make it possible to use the same pin for vision in and output (DIN-AV). You only have to change the biasing to make it possible to drive these inputs. No cost and no room needed.

Kind regards
Darius

[Heatercathodeshort]

Hello,
I have just purchased an AURORA converter and all I can say is that it produces a VERY good picture reproducing the 3mc/s bars easily on my 14" GEC BT1748 1956 receiver. It will match the quality of the original BBC signal and its about as good as you can get. I'm happy! Regards John.

[oldeurope]

Anyone able to show us a screenshot from the spectrum analyzer? (modulator set to London ch1)

Darius

[tubesrule]

Originally Posted by oldeurope

Good morning,
the vision output will be used to record something on for example vhs tapes or to drive another modulator. Often old vcrs are used and they sometimes have a cap and resistor terminated input. This design is used to make it possible to use the same pin for vision in and output (DIN-AV). You only have to change the biasing to make it possible to drive these inputs. No cost and no room needed.

Kind regards
Darius

Hi Darius,
While this will potentially help, it does pose some engineering challanges. Increasing the load on the transistor will increase the base current which needs to be accounted for in the D/A section. Also, this will move the hfe curve further into a non-linear region which will affect the output. I've attached the schematic of this section to illustrate.

Originally Posted by oldeurope

Anyone able to show us a screenshot from the spectrum analyzer? (modulator set to London ch1)

I can do this Darius. What specifically are you looking for? (Span, RBW, video content?)

Take care,
Darryl

[tubesrule]

Oops, forgot the attachment.

Darryl

[ppppenguin]

In my review of the Aurora I mentioned:

.... a novel discrete DAC....

Now Darryl has published the schematic you can see the subtlety for yourselves. At first sight it looks like a conventional R-2R ladder DAC but the resistor values are a bit strange and this is a 10 bit DAC. I sort of half guessed how it worked and the user manual has a small clue about pulse width modulation. Darryl explained it to me properly and I was very impressed.

Have a look and judge for yourselves.

[oldeurope]

Originally Posted by tubesrule

Hi Darius,
While this will potentially help, it does pose some engineering challanges. Increasing the load on the transistor will increase the base current which needs to be accounted for in the D/A section. Also, this will move the hfe curve further into a non-linear region which will affect the output. I've attached the schematic of this section to illustrate.

I can do this Darius. What specifically are you looking for? (Span, RBW, video content?)

Take care,
Darryl

Good morning Darryl, thanks for the schematic. I am impressed what can be done in one IC. My analogue design is much more complicated but discrete.
You are using an R-2R D/A section. The load is not increased by changing the biasing of Q1. The base current is not importand for the R-2R and the filter.
The input resistance of the emitter follower does not change with the biasing because R36/7/8 are still the same!
It is 40K Ohms without the biasing resistors R34/35.
Vice versa you are getting into a more linear region with a higher biasing voltage because the ratio of AC current changes to DC current of the transistor gets lower by higher biasing. With your biasing the non linearity is most at the sync (tips). They are square wave and the sync will be made a bit shorter at the output. You can take every value for R34/35 you want, if you make sure that R34 II R35 does not change.

Attached some screenshots from the analyzer with my modulator without filters at the output. Input is a colour bar without colour.

Kind regards
Darius

[ppppenguin]

I hope this is correct. The output impedance of a R-2R ladder DAC is R (in this case 1K) regardless of the digital code. I don't think the departaure from the pure R-2R structure affects this. There are very small variations with code due to the finite output impedance of the digital drives to the DAC. These are not important.

Hence the Thevenin model of the DAC is a variable voltage source (0-3.3V) in series with a 1K resistor. The resistive load on the DAC affects only the gain, not the linearity.

The base current of Q1 is very close to Ie/HFE (large signal parameters). This will vary between devices and with temperature. The input impedance of Q1 is Re/hfe (total load on emitter divided by small signal current gain). The signal is actually large in this application so hfe will vary with the signal and cause non-linearity.

I think that the resistance of R34//R35 will dominate the load at the DAC output so the hfe variations will not cause any significant non-linearity.

[oldeurope]

Originally Posted by ppppenguin

I hope this is correct. The output impedance of a R-2R ladder DAC is R (in this case 1K) regardless of the digital code. I don't think the departaure from the pure R-2R structure affects this. There are very small variations with code due to the finite output impedance of the digital drives to the DAC. These are not important.

Hence the Thevenin model of the DAC is a variable voltage source (0-3.3V) in series with a 1K resistor. The resistive load on the DAC affects only the gain, not the linearity.

The base current of Q1 is very close to Ie/HFE (large signal parameters). This will vary between devices and with temperature. The input impedance of Q1 is Re/hfe (total load on emitter divided by small signal current gain). The signal is actually large in this application so hfe will vary with the signal and cause non-linearity.

I think that the resistance of R34//R35 will dominate the load at the DAC output so the hfe variations will not cause any significant non-linearity.

Good morning Jeffrey,
I totally agree.
R34//R35 (// means parallel) sets the output level.
A higher biasing makes the non linearity "unmeasureable".

Kind regards
Darius

BTW The output level at VID_CONV changes a little bit with the load.
It is visible! It is better to take a separate emitter follower for each output.
(It is the same with opamps!)

[ppppenguin]

Originally Posted by oldeurope

It is better to take a separate emitter follower for each output. (It is the same with opamps!)

You are correct. The output amplitude will change with loading due to finite output impedance. Separate emitter followers are better but in this case Darryl took the decision to use a single emitter follower.

It is NOT the same with opamps. The closed loop output impedance of an opamp is very low, much lower than an emitter follower. In video applications I use opamps to drive multiple 75R loads, each output via its own 75R resistor. Some opamps are better than others for driving several loads. The EL2x44 series will drive 2 loads, the older EL2020 will drive 3, the LM6172 will drive at least 4. The interaction between outputs is almost unmeasurable (<<0.1dB) at low frequencies and may increase up to 1dB at 10MHz depending on the opamp. For the highest quality applications the maximum loading is usually set by the rise in differential gain and differential phase.

[oldeurope]

Originally Posted by ppppenguin

...It is NOT the same with opamps. The closed loop output impedance of an opamp is very low, much lower than an emitter follower. In video applications I use opamps to drive multiple 75R loads, each output via its own 75R resistor. Some opamps are better than others for driving several loads. The EL2x44 series will drive 2 loads, the older EL2020 will drive 3, the LM6172 will drive at least 4. The interaction between outputs is almost unmeasurable (<<0.1dB) at low frequencies and may increase up to 1dB at 10MHz depending on the opamp. For the highest quality applications the maximum loading is usually set by the rise in differential gain and differential phase.

Hi Jeffrey, I know this from the commercially available video splitters.
You always "see" something if something is done on an other output.
(With difficult loads, some start to oscillate This is the kind with 1Vpp and 0R or a coil (?) at the output.)
Good splitters have an opamp for each output.

Another Question:
I had a look at the schematic.
There are seven bits generating the vision signal. Means 128 steps.
Sync takes 30% means there are only 90 grey steps left for the vision.
Am I right?

Kind regards
Darius

[ppppenguin]

Originally Posted by oldeurope

I know this from the commercially available video splitters.You always "see" something if something is done on an other output. (With difficult loads, some start to oscillate This is the kind with 1Vpp and 0R or a coil (?) at the output.)

Good splitters have an opamp for each output.

When correctly designed you need only 1 opamp. I have designed very high quality equipment with multiple video outputs to be used by broadcasters. I have never used one opamp per output. The isolation between outputs is always good. You can even put a long unterminated length of cable or a short circuit on 1 output and the effect on the others will be very small. The 75R resistor in series with each each output ensure that the load on the opamp is always mainly resistive.

Originally Posted by oldeurope

Another Question:
I had a look at the schematic.
There are seven bits generating the vision signal. Means 128 steps.
Sync takes 30% means there are only 90 grey steps left for the vision.
Am I right?

No. This is a 10 bit DAC! VDO9-VDO6 are the 4 MSBs as in a normal R-2R DAC. VDO4/5, VDO2/3 and VDO0/1 are pulse width modulated signals and represent 2 bits each. The output is sampled at 3 times the normal speed. Here are the pulse width modulated waveforms:

00 ____________

01 ____--____--

10 __----__----

11 ------------

These waveforms are integrated by the filter to give the correct average values. The resistor values are different for these lower bits. You have to sit down and do a lot of calculation (mainly using Thevenin's theorem) to prove how this works.

I have designed pulse width modulated DACs but only for very low speed control systems. In one design I use a 12 bit PWM DAC that requires only 1 pin on an FPGA to adjust the fine tuning on a TCXO. Darryl's design is a hybrid between R-2R and PWM and is very clever. It is probably more accurate than using extra resistors. The main reason he did this design was shortage of pins on the FPGA.

[oldeurope]

Originally Posted by ppppenguin

When correctly designed you need only 1 opamp. I have designed very high quality equipment with multiple video outputs to be used by broadcasters. I have never used one opamp per output. The isolation between outputs is always good. You can even put a long unterminated length of cable or a short circuit on 1 output and the effect on the others will be very small. The 75R resistor in series with each each output ensure that the load on the opamp is always mainly resistive.



No. This is a 10 bit DAC! VDO9-VDO6 are the 4 MSBs as in a normal R-2R DAC. VDO4/5, VDO2/3 and VDO0/1 are pulse width modulated signals and represent 2 bits each. The output is sampled at 3 times the normal speed. Here are the pulse width modulated waveforms:

00 ____________

01 ____--____--

10 __----__----

11 ------------

These waveforms are integrated by the filter to give the correct average values. The resistor values are different for these lower bits. You have to sit down and do a lot of calculation (mainly using Thevenin's theorem) to prove how this works.

I have designed pulse width modulated DACs but only for very low speed control systems. In one design I use a 12 bit PWM DAC that requires only 1 pin on an FPGA to adjust the fine tuning on a TCXO. Darryl's design is a hybrid between R-2R and PWM and is very clever. It is probably more accurate than using extra resistors. The main reason he did this design was shortage of pins on the FPGA.

Hi Jeffrey,
thanks for explaining this. It is new for me. So there are 700 grey steps.
This is good!
To compare it with an analogue world: The CXL5506 has typical 56dB s/n
things are visible more than 6db below noise makes 56dB+ 6dB say 60dB
this is 1000 minus 30% (sync) gives 700, too.

BTW: 0,1dB is 1% this is definately visible!

Kind regards
Darius

[tubesrule]

Hello Darius and Jeff,
All the discussion about the output driver seems spot on. There are non-linear effects caused by the follower, but regardless of the load, they are minimal. The output impedance of the D/A is 1K as Jeff pointed out, regardless of the digital code or of the modified nature of the pwm bits. R34, the 2.32K, is installed (R35 is not) so the effective D/A output impedance is about 700R. Both R34 and R35 are never installed at the same time. Only one is used to set the D/A gain and operating point.

The hfe curve for the BC848 is the flatest between approximately 1-30ma which is where it is biased to operate currently. Beyond 30ma the hfe has a pronounced rolloff that would result in a change in gain for the video portion of the signal. I believe Darius correctly pointed out that the sync portion of the signal is most affected by the low end of the hfe curve now, but pushing the biasing beyond 30ma would mean the video portion would be affected. BTW, the sync signals generated are not just square waves but are edge rate controled as required.

Darius is correct that independent followers would be ideal, however this isn't required. While there will be a gain change to the signal fed to the internal modulator (VID_CONV) when an external load is connected, it is negligable and stable if the load is the correct 75R.

I have also run several video outputs through 75R terminators from a single op-amp output on broadcast designs. Along with the op-amps mentioned by Jeff, the Linear Tech LT12XX series among others can provide excellent isolation with nearly immeasurable effect from one output to another. This type of output was considered for this design, but the slight advantages in no way outweighed the cost increase.

The D/A section is effectively 10 bit, and will achieve a better than 56dB snr with the post filter as described on the schematic. Using the current 2.32K for R34, the D/A uses 942 steps for the entire video signal with sync. One of the current test modes on the unit allows the validation of the snr by applying a worst case bit pattern to the D/A.

Darryl

[oldeurope]

Thanks Darryl.
One thing I don't understand.
At peak white all bits are high (=3.3V). So there is 3.3V at base of Q1.
2.6V at its emitter. The sync tips are 2Volts lower this is 0.6V.
Why do you have 0.2 Volts there?

Kind regards,
Darius

[ppppenguin]

Darryl said that peak white is at code 942, not 1023. This is approximately 3.3 * 942/1024 = 3V so the sync tip will be at 1V at the base. Around 0.3V at the emitter.

I'm not sure why Darryl chose white level at 942. In professional digital composite video, white level is 844. This is because the subcarrier in saturated colours rises well above white. 942 allows a good margin for signals that rise above white. This is good practice since clipped whites usually look nasty. I have put white level at 853 in one design because this gave black level at 256 which was convenient.

[oldeurope]

Originally Posted by ppppenguin

Darryl said that peak white is at code 942, not 1023. This is approximately 3.3 * 942/1024 = 3V so the sync tip will be at 1V at the base. Around 0.3V at the emitter.

I'm not sure why Darryl chose white level at 942. In professional digital composite video, white level is 844. This is because the subcarrier in saturated colours rises well above white. 942 allows a good margin for signals that rise above white. This is good practice since clipped whites usually look nasty. I have put white level at 853 in one design because this gave black level at 256 which was convenient.

Ah, thanks I understand.

Darius

[tubesrule]

Hi,
It's sort of straight forward where the 942 comes from. As Jeff mentioned, 844 is the typical white point value. This is required for a normal composite signal since the chroma information can go much higher. When only a monochrome output is required, you can take advantage of more of the D/A range. Since 0-100IRE represents 16-235, or 220 steps inclusive in a standard ITU656 data stream, and using a 70/30 video/sync ratio, the sync occupies approximately 94 steps for a total of 314. Since all the interpolation math is done at 12 bits precision, I round the results to the nearest whole integer multiple that best utilizes the 10 bit D/A range. This turns out to be a multiplier of 3 which results in the 3 * 314 = 942 steps. This yields 942/1024 * 3.3V = 3.03V range out of the D/A. If I had to add composite chroma, this would need to be dropped to the more conventional 844. With the 942 number, it allows for almost 10IRE of super white in the video.
This digital range is then matched to the output stage using the set resistors R34 and R35. Between how the digital range and set resistors are calculated, you can control the gain and operating point.

I've included a couple of picts from the spectrum analyzer of the RF output.

Darryl