Continue to Site

Welcome to EDAboard.com

Welcome to our site! EDAboard.com is an international Electronics Discussion Forum focused on EDA software, circuits, schematics, books, theory, papers, asic, pld, 8051, DSP, Network, RF, Analog Design, PCB, Service Manuals... and a whole lot more! To participate you need to register. Registration is free. Click here to register now.

hardware design for gaussian filter

Status
Not open for further replies.

suribright

Member level 2
Joined
Oct 6, 2011
Messages
42
Helped
0
Reputation
0
Reaction score
0
Trophy points
1,286
Activity points
1,576
HI all,
Any one know about the gaussian filter of gmsk. I am looking for the hardware design not a simulation. Please let me know. Thanks
 

An analog gaussian filter can be obtained as a cascade of identical filters.
You can construct a low-pass gaussian filter cascading several single-pole identical filters.
The number of sections determines how similar to a true gaussian characteristic is the result.
Regards

Z
 

Hi Zorro, I have two questions. First that lowpass gaussian filter there is the one in gmsk gsm or not?. Second, the lowpass you said there is RC? I have already added several RC-RC-RC but the negative side still exit. I may use active lowpass instead.
 

Hi Zorro, I have two questions. First that lowpass gaussian filter there is the one in gmsk gsm or not?. Second, the lowpass you said there is RC? I have already added several RC-RC-RC but the negative side still exit. I may use active lowpass instead.

A Gaussian filter has complex poles - thus, for passive realization you need RLC groups or you must consider an active RC solution.
 

So that active RC solution is the one used in GSM GMSK or not?. I am not sure of that. Is there anyone know about GSM gaussian filter?. I could not find it.
 

So that active RC solution is the one used in GSM GMSK or not?. I am not sure of that. Is there anyone know about GSM gaussian filter?. I could not find it.

For realizing complex poles you always have the choice between a passive RLC or an active RC solution. It depends on some other (environmental) conditions like power supply availability, weight, space, external load,...
 

Gaussian filter implementations are aproximations, because an ideal gaussian filter isn't causal. As far as I see, a cascade of buffered first order low-passes can work as a gaussian approximation as well, but I doubt that GMSK filters are using it. Active filters with low Q complex pole pairs can achieve the aproximation with considerably lower effort.

P.S.: The GSM modulation is specified in GSM 05.04, latest document version ETSI EN 300 959 V8.1.2 (2001-02), downloadable at

As far as I understand, the standard specifies an ideal gaussian filter with tolerances, in so far the implementation details are up to you.
 
Last edited:
  • Like
Reactions: LvW

    LvW

    Points: 2
    Helpful Answer Positive Rating
Hi FvM. Assume I have a pulse with a width of 10us. If I want to create gaussian pulse, I use several RC-RC filter (150ohm-10ns) to shape that pulse to gaussian pulse. But the shape is not 100% look like (see picture). The falling edge of yellow signal always takes at the falling of pulse. I would like to have gaussian pulse which the shape inside the pulse. I believe that GSM one is in the middle . SO do you know how shape that in the middle and how to shape this rec_pulse to get gaussian? . Tks. **broken link removed**
 

If I want to create gaussian pulse, I use several RC-RC filter (150ohm-10ns) to shape that pulse to gaussian pulse. But the shape is not 100% look like (see picture).

As mentioned already, an ideal filter with Gaussian response cannot be realized (exponential function). It has to be approximated by a "broken-rationale" function.
A simple series combination of RC elements has only real poles and leads to a lowpass with "critical damping" - however, with rising filter degree (number of RC units) a Gaussian response in the time domain can be approximated.
 

"Invalid attachment", unfortunately I can't comment your waveforms.
 

Hi FvM, Here is the attached file. Hope you can see it now. **broken link removed**https://www.edaboard.com/attachment.php?attachmentid=80077 . @LvW, please explain more about " rising fillter degree " I do not quite understand this term. Is that the simple series RC?. In that case designing the perfect gaussian pulse will not be possible. It is just approximately. The pulse signal through the RC always have the rising exponential and then the falling exponential. The falling exponential part does not look like in the gausian. Even with series of RC still exist this part some how. I read that the gmsk has perfect gaussian (simulation only). Do not know how they design in practical?.
 

@LvW, please explain more about " rising fillter degree " I do not quite understand this term. Is that the simple series RC?. In that case designing the perfect gaussian pulse will not be possible. It is just approximately. The pulse signal through the RC always have the rising exponential and then the falling exponential.

* Rising degree of the filter means: Increasing number of RC sections
* My remark (exponential response) was related to the frequency domain (transfer function) - not to the time domain (step response).
 

@FvM. I think some things wrong with your browser. I used firefox instead and I can see the picture. Please use another browser. @ LvM. Yes, I have increased number of RC in series. The gaussian does not look exactly as perfect gaussian pulse. The time is also longer than the rec_pulse. So from what you said we cannot have a perfect gaussian pulse isnt it?.
 

@FvM. I think some things wrong with your browser. I used firefox instead and I can see the picture. Please use another browser.
Curiously your attachments are two out of >10000 edaboard attachments that display correctly. They neither display in firefox.

Other users manage somehow to post their attachments in a readable way. There may be specific edaboard problem with your attachment I'm not aware of. But I have no means to solve it.

Referring to gaussian filter problem: Do you use decoupled (buffered) first order filters or just an RC ladder? This would be in fact the least effective way to approximate a gaussian filter, as mentioned before. Decoupled (buffered) first order RC is better but still ineffective. A higher order filter with complex pole pairs is the preferred way to go.

- - - Updated - - -

Please see impulse responses of third and fifth order complex gaussian filter approximations

 
  • Like
Reactions: LvW

    LvW

    Points: 2
    Helpful Answer Positive Rating
@ LvM. Yes, I have increased number of RC in series. The gaussian does not look exactly as perfect gaussian pulse. The time is also longer than the rec_pulse. So from what you said we cannot have a perfect gaussian pulse isnt it?.

As mentioned several times - an exact gaussian behaviour is not possible! It only can approximated.
 

@FvM: Yep, I used insert image button and insert attachment in. Dont know why it is not work in your case. I used normal RC ladder only and I think your higher order filter is better than mine. Would it possible you post your circuit please?. I would like to see that. Thanks.
@LvM: So in GSM they have used gaussian filter for gmsk. This will attract bit 1 or 0 before modulating. If gaussian filter is not perfect the data may not be corrected I think. Dont know how gaussian is used in real gsm

- - - Updated - - -

Also I have used LTspice to simulate all the circuit like this. Do you guy know any others?. which one is the best? Tks
 

The filters are just 3th respectively 5th order low-passes that can be implemented in any topology like passive LC or popular active filters (MFB, Sallen Key).

I don't know the actual GSM bandwith, thus I show a normalized 5th order Sallen Key filter (with "absorbed" 1st order) , as in the time domain waveform.


If gaussian filter is not perfect the data may not be corrected I think. Dont know how gaussian is used in real gsm
As verbosely explained, a real gaussian filter won't be "perfect". I only took a brief look at GSM 05.04 and 05.05. As far as I understand, the filter performance is specified as maximum modulation phase error (and the modulation spectrum mask on the other hand). Something you most likely need to calculate if you don't have other literature deriving the filter specification explicitely.
 
Last edited:
  • Like
Reactions: LvW

    LvW

    Points: 2
    Helpful Answer Positive Rating
Ok. thank FvM and LvW for your discussion. I will try this high order filter to see how it goes. However, the input signal is the pulse not the sin as in your graph. I need to filter the pulse to become gaussian.
 

Status
Not open for further replies.

Similar threads

Part and Inventory Search

Welcome to EDABoard.com

Sponsor

Back
Top