• posted

Dear All,

I need to design an atteuator to simulate 9000ft of ADSL cable loss. Could anybody please let me know what are are the do's and don't's on practical design of it. Based on my studies, it should give 22.76 dB of attenuation at 150 kHz and 50.7455dB attenuation at 772 kHz. What else should i considered more?

Best Regards

• posted

test -------------------------------------------

• posted

I assume you are talking about an actual hardware attenuator, and not just a software simulation?

With practical transmission line simulation like this I've always found it's prudent to actually use the real cable in a real system test. It would not be out of the bounds of practicality to actually use 9000ft of cable (I've done this sort of length before a few times), or test a shorter length and extrapolate your results. You can also test the performance of a real cable and match it with any "simulated" attenuator.

Dave.

• posted

This reminds me of something I made in the late '90s. I had to test HDSL cards and we didn't like the idea of heavy cable drums in our manufacturing test setup. So I made some equivalent circuits out of discrete R,L and C components on a board that could plug into our racks.

Steps:

1. Get the RLGC parameters for the line you want to simulate. These will be a function of frequency. Bellcore used to publish technical reports with cable parameters measured on a representative sample of actual subscriber lines.

1. Work out how accurate your simulation needs to be, in terms of dB error as a function of frequency. There will probably be some upper frequency, beyond which you don't care about the accuracy at all (which is good, as most simulations based on lumped elements have an upper frequency limit beyond which the accuracy gets really bad). BTW, your use of so many decimal places in 22.76 and 50.7455 dB is rather naive; you are unlikely to get an accuracy of better than +/- a few dB over any non-trivial frequency range.

2. Create a lumped model of the line (see the links (below) for ideas), and check it in your preferred version of Spice. Increase the number of sections (which reduces the ripple and increses the upper frequency limit) until you get the required accuracy with practical component values. Check both the attenuation and impedance. The inductors and (to a lesser extent) the capacitors will not behave like their ideal models. You will need to include this in your simulation. BTW, avoid ceramic caps here, unless they are NP0.

The loss in these reactive components actually helps with the simulation. Remember, we are trying to recreate the behaviour of something that has both skin effect and dielectric losses.

1. Lay it out on a PCB. There will probably be a few hundred sections, so a PCB is the only practical way. You must be very careful with the layout, as crosstalk will ruin the attenuation characteristics. (BTW, crosstalk is also a problem for cable drums.)

1. Build and test.

d/ec84b4ce3d5a07c3

fb52/

Regards, Allan

• posted

What are you trying to prove by this simulation?

The main problem is the direct capacitive coupling between the input and the output. Because of that, the frequency/phase response can be very peculiar albeit have nothing to do with the reality. Do not use a piece of actual cable cable winded on a bobbin; you will have to unwind it.

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

• posted

My thoughts on this may be hopelessly naieve but I'd think about doing it as two 10dB H-pads and a low pass filter that rolls off at about 13dB per octave.

I suspect that it needs to be more or less symetrical as you want to communicate both ways.

Inductor and capacitor filter in the middle of the circuit, 10dB pads at both ends. Having the pads at the ends will make the source impedance look more like the characteristic impedance of the line which is probably around 110ohms but you need to look that up yourself.

There are a few h-pad calculators on the web eg

I'd exepct you could come up with a filter design with about eight elements that would give a suitable rolloff. Might need quite large inductors due to the low frequency.

Bob

• posted

Sure he did, did you not see the question marks?

• posted

We did a G.SHDSL MODEM design for a non-telecom application. We decided that we wanted to test with real UTP in various lengths, and we wanted to do so not only during development, but also automatically during production. Our requirement was operation up to 5km, but I designed the test for 6km.

Using spools of wire is not an option for obvious reasons. This test requires a non-inductive (zig-zag) placement of the cable. In order to keep this manageable, we used off-the-shelf low-cost 25-pair telecom cables with telecom connectors. We jig-jagged the cables through the roof trusses in our industrial unit.

I designed a small PCB to be placed in the chain between 25-pair cables, and at the ends of the chain. The PCBs had two main functions. The first was to loop back pairs on the cable, to create a

25-leg zig-jag along any given cable. The second was to provide a relay that would either continue the signal pair to the next cable, or tap off to go back to the test set.

Using this arrangement the tester software can automatically select any one of the following lengths:

100? / 30m 1,350? / 411m 2,600? / 792m 7,600? / 2,316m 12,600? / 3,840m 17,600? / 5,364m 20,100? / 6,126m

The initial 100' foot length is actually two 1-pair 50' lengths running between the relay boards fixture and the test set.

================================

Greg Neff VP Engineering

*Microsym* Computers Inc. snipped-for-privacy@guesswhichwordgoeshere.com
• posted

Dear all,,

Now I have better confidence for starting up.

By the way... Mr Allan...I cannot reach the links you described there..somehow they are not complete...could you please provide the title of the discussions??

Best Regards..

SL

,
f

=A0

ns,

• posted

Yes Mr Dave..

I am talking about an actual hardware attenuator.

From practical measurement I found out that the series resistance value is 250 ohms in each wire. So I am thinking of adding series resistance of 250 ohms to simulate the resistance of the wire.

---/\\/\\/\\-----

.../\\/\\/\\....

This will give an attenuation of -3dB for all frequencies.

Then I am thinking of creating a T network with two series inductors and one parallel capacitor which will give attenuation of around -20dB at 150kHz.

Then I will addjust values acordingly to have the attenutaion around

-51dB at 772 kHz.

Please correct me if you find any practically impossible things in my idea...

I am new in this kind of things..

Thank you so much for everything..

Regards

• posted

Think of it as an aptitude test.

Hint: find my post at groups.google.com, click on "more options" then click on "show original".

Allan

• posted

ck

Thank you so much.

Now I am able to find it.

Regards

• posted

One way to design such a simulator is to use optimization. The program CCICAP has a genetic algorithm based optimization feature that is well suited to this type of design.

I have included Myauk's specifications in an example that can be studied here:

I used two identical sections of RLC low pass filters. Using guessed values for the RLC components results in an initial error at the specified frequencies of 37.6 dB. After optimization the error is reduced to .08 dB.

The example could be improved upon with more sections if additional attenuation specifications are available.

Regards, Bob Smither

• posted

They do sell DSL loop simulators by the way. Different models, some with and without noise sources and programmable loop lengths to allow automated testing.

• posted

I think they have more features than it is necessary for us. Do you know any good source where I could check the price? If the cost is less than the cost of developing it myself, I'd better buy it. :P

• posted

Dear Allan..

=A0Bellcore used to publish technical

Is there any link to this? I could not find it.

=A0BTW, avoid ceramic caps here, unless they are NP0.

Could you please explain to me why I should avoid using ceramic caps?

Regards

• posted

We looked at a G.SHDSL wireline simulator from Spirent. If I remember correctly, the cost of one of these was about the same as a new Lexus ES350.

In our case we did not need the impairment generation features. All we were required to do was verify operation over some lengths up to and including 5km.

It was easier for us to use real cable, because if we used a simulator (especially home-grown) then we would have had to verify that the simulator was equivalent to cable.

================================

Greg Neff VP Engineering

*Microsym* Computers Inc. snipped-for-privacy@guesswhichwordgoeshere.com
• posted

In one of the posts I directed you to, I mentioned these standards:

ETSI ETR-152 (HDSL standard) ETSI ETR-080 (ISDN U interface standard) ANSI T1.601 (ISDN U interface standard)

Ceramic caps have a "voltage coefficient" that describes their change in capacitance with the applied voltage. This will produce non-linear distortion of your signals, and reduce the accuracy of your smimulation. The wires you are trying to model typically use some sort of plastic insulation (e.g. polyethylene) which does not have this effect.

BTW, you should also make sure the inductors are used well below saturation, for similar reasons.

Regards, Allan

• posted

So ceramic caps and saturated inductors generate sidebands?

(...Runs away as everyone thows rotten fruit...)

• posted

:Dear All, : :I need to design an atteuator to simulate 9000ft of ADSL cable loss. :Could anybody please let me know what are are the do's and don't's on :practical design of it. Based on my studies, it should give 22.76 dB :of attenuation at 150 kHz and 50.7455dB attenuation at 772 kHz. :What else should i considered more? : :Best Regards

I remembered seeing some artificial line section values in a document I was researching recently. You can ind this document here