On Wed, 24 Oct 2012 07:58:18 -0700, Joerg <invalid@invalid.invalid> wrote:>krw@att.bizzzzzzzzzzzz wrote: >> On Tue, 23 Oct 2012 17:49:19 -0700, Joerg <invalid@invalid.invalid> wrote: >> >>> krw@att.bizzzzzzzzzzzz wrote: >>>> On Tue, 23 Oct 2012 14:19:16 -0700, Joerg <invalid@invalid.invalid> wrote: >>>> >>>>> krw@att.bizzzzzzzzzzzz wrote: >>>>>> On Tue, 23 Oct 2012 12:45:14 -0700, Joerg <invalid@invalid.invalid> wrote: >>>>>> >>>>>>> krw@att.bizzzzzzzzzzzz wrote: > >[...] > >>> I do use others, just not for really critical stuff like this where >>> simulation is key. >> >> With a behavioral, you can only simulate what the model builder anticipated. >> Again, that's rarely where the demons be. At least with full models, you have >> a chance. >> > >Yep. Last time I did that one single run too about four (!) hours.<shrug> I've used entire mainframes for weeks on a single simulation. CPUs weren't as cheap, nor as powerful, as they are now. Not everyone had one sitting on their desk, either.>>>>> For some of my clients this is relevant. I have used PWM chips in the >>>>> past that are 2nd-sourced. From Richtek and TI, for example. But those >>>>> tend to be older designs. >>>> Jellybeans, perhaps. Nothing leading edge. >>>> >>>>>>> Some >>>>>>> don't have current-mode control and that's a no-no here. >>>>>> Almost all do. Everything I've used is current-mode. >>>>>> >>>>> I am replacing a TI regulator that isn't, and this is one of the reasons. >>>> Are you saying5 that TI doesn't have current mode regulators? <boggle> Almost >>>> everything I've looked at in the last year is current mode. >>>> >>> Huh? That's not what I said. I said I am replacing a TI regulator that >>> is not current-mode. It has to go because that one isn't up to the job. >>> Of course they have developed better ones by now. But no behavioral >>> models so one cannot simulate in a reasonable time frame. >> >> Yet you will only use LTC because they're current mode. <boggle> >> > >You are not correctly reading what I write. I am using LTC because there >are models that allow fast simulation. _Not_ because they are current >mode. I only require that any switcher that replaces the old one is >current mode.That's not what you said. You said it like current mode was something unique. It surely isn't.>>>>>>> So far pretty >>>>>>> much all my designs came out almost verbatim as simulated. That's the >>>>>>> kind of time-to-market clients want. >>>>>> You trust behaviorals? Amazing. >>>>> Depends on who made them. From Linear Technology, yes, usually I trust >>>>> them. Many years of successful design with their models gives me confidence. >>>> I sure don't. I've never seen a behavioral that's good outside it's *NORMAL* >>>> operating conditions. That's never where the demons are. >>>> >>> Then you haven't tried hard enough with LTSpice :-) >> >> Nonsense. >> >>> By now they know me at LTC for lots of unorthodox uses. >> >> The models are useless outside their limited operational area. IOW, useless >> for guaranteeing operation. >> > >Sorry, but I have proven that statement wrong many times. For that to >work you do need to establish good connections to their engineers >though. I have.YOu *think* you've simulated everything that can go wrong. Obviously a model can't be used outside it's limits. Behaviorals have very limited useful area and they're rarely documented.
Large capacitance varicaps, where are thee?
Started by ●October 22, 2012
Reply by ●October 24, 20122012-10-24
Reply by ●October 24, 20122012-10-24
krw@att.bizzzzzzzzzzzz wrote:> On Wed, 24 Oct 2012 07:58:18 -0700, Joerg <invalid@invalid.invalid> wrote: > >> krw@att.bizzzzzzzzzzzz wrote: >>> On Tue, 23 Oct 2012 17:49:19 -0700, Joerg <invalid@invalid.invalid> wrote: >>> >>>> krw@att.bizzzzzzzzzzzz wrote: >>>>> On Tue, 23 Oct 2012 14:19:16 -0700, Joerg <invalid@invalid.invalid> wrote: >>>>> >>>>>> krw@att.bizzzzzzzzzzzz wrote: >>>>>>> On Tue, 23 Oct 2012 12:45:14 -0700, Joerg <invalid@invalid.invalid> wrote: >>>>>>> >>>>>>>> krw@att.bizzzzzzzzzzzz wrote: >> [...] >> >>>> I do use others, just not for really critical stuff like this where >>>> simulation is key. >>> With a behavioral, you can only simulate what the model builder anticipated. >>> Again, that's rarely where the demons be. At least with full models, you have >>> a chance. >>> >> Yep. Last time I did that one single run too about four (!) hours. > > <shrug> I've used entire mainframes for weeks on a single simulation. CPUs > weren't as cheap, nor as powerful, as they are now. Not everyone had one > sitting on their desk, either. >Well, I do not have a mainframe available here.>>>>>> For some of my clients this is relevant. I have used PWM chips in the >>>>>> past that are 2nd-sourced. From Richtek and TI, for example. But those >>>>>> tend to be older designs. >>>>> Jellybeans, perhaps. Nothing leading edge. >>>>> >>>>>>>> Some >>>>>>>> don't have current-mode control and that's a no-no here. >>>>>>> Almost all do. Everything I've used is current-mode. >>>>>>> >>>>>> I am replacing a TI regulator that isn't, and this is one of the reasons. >>>>> Are you saying5 that TI doesn't have current mode regulators? <boggle> Almost >>>>> everything I've looked at in the last year is current mode. >>>>> >>>> Huh? That's not what I said. I said I am replacing a TI regulator that >>>> is not current-mode. It has to go because that one isn't up to the job. >>>> Of course they have developed better ones by now. But no behavioral >>>> models so one cannot simulate in a reasonable time frame. >>> Yet you will only use LTC because they're current mode. <boggle> >>> >> You are not correctly reading what I write. I am using LTC because there >> are models that allow fast simulation. _Not_ because they are current >> mode. I only require that any switcher that replaces the old one is >> current mode. > > That's not what you said. You said it like current mode was something unique.Where?> It surely isn't. > >>>>>>>> So far pretty >>>>>>>> much all my designs came out almost verbatim as simulated. That's the >>>>>>>> kind of time-to-market clients want. >>>>>>> You trust behaviorals? Amazing. >>>>>> Depends on who made them. From Linear Technology, yes, usually I trust >>>>>> them. Many years of successful design with their models gives me confidence. >>>>> I sure don't. I've never seen a behavioral that's good outside it's *NORMAL* >>>>> operating conditions. That's never where the demons are. >>>>> >>>> Then you haven't tried hard enough with LTSpice :-) >>> Nonsense. >>> >>>> By now they know me at LTC for lots of unorthodox uses. >>> The models are useless outside their limited operational area. IOW, useless >>> for guaranteeing operation. >>> >> Sorry, but I have proven that statement wrong many times. For that to >> work you do need to establish good connections to their engineers >> though. I have. > > YOu *think* you've simulated everything that can go wrong. Obviously a model > can't be used outside it's limits. Behaviorals have very limited useful area > and they're rarely documented. >The same goes for full models. There are limits to everything virtual. However, with behavioral models I've had years of successful designs. And yeah, I did discover the occasional model bug and also the occasional real bug on an IC. Where sim and real life did not jibe. Usually it meant that the datasheet didn't either. That's part of life. -- Regards, Joerg http://www.analogconsultants.com/
Reply by ●October 24, 20122012-10-24
On Wed, 24 Oct 2012 10:33:46 -0700, Joerg <invalid@invalid.invalid> wrote:>krw@att.bizzzzzzzzzzzz wrote: >> On Wed, 24 Oct 2012 07:58:18 -0700, Joerg <invalid@invalid.invalid> wrote: >> >>> krw@att.bizzzzzzzzzzzz wrote: >>>> On Tue, 23 Oct 2012 17:49:19 -0700, Joerg <invalid@invalid.invalid> wrote: >>>> >>>>> krw@att.bizzzzzzzzzzzz wrote: >>>>>> On Tue, 23 Oct 2012 14:19:16 -0700, Joerg <invalid@invalid.invalid> wrote: >>>>>> >>>>>>> krw@att.bizzzzzzzzzzzz wrote: >>>>>>>> On Tue, 23 Oct 2012 12:45:14 -0700, Joerg <invalid@invalid.invalid> wrote: >>>>>>>> >>>>>>>>> krw@att.bizzzzzzzzzzzz wrote: >>> [...] >>> >>>>> I do use others, just not for really critical stuff like this where >>>>> simulation is key. >>>> With a behavioral, you can only simulate what the model builder anticipated. >>>> Again, that's rarely where the demons be. At least with full models, you have >>>> a chance. >>>> >>> Yep. Last time I did that one single run too about four (!) hours. >> >> <shrug> I've used entire mainframes for weeks on a single simulation. CPUs >> weren't as cheap, nor as powerful, as they are now. Not everyone had one >> sitting on their desk, either. >> > >Well, I do not have a mainframe available here.You have about 100x the CPU power there.>>>>>>> For some of my clients this is relevant. I have used PWM chips in the >>>>>>> past that are 2nd-sourced. From Richtek and TI, for example. But those >>>>>>> tend to be older designs. >>>>>> Jellybeans, perhaps. Nothing leading edge. >>>>>> >>>>>>>>> Some >>>>>>>>> don't have current-mode control and that's a no-no here. >>>>>>>> Almost all do. Everything I've used is current-mode. >>>>>>>> >>>>>>> I am replacing a TI regulator that isn't, and this is one of the reasons. >>>>>> Are you saying5 that TI doesn't have current mode regulators? <boggle> Almost >>>>>> everything I've looked at in the last year is current mode. >>>>>> >>>>> Huh? That's not what I said. I said I am replacing a TI regulator that >>>>> is not current-mode. It has to go because that one isn't up to the job. >>>>> Of course they have developed better ones by now. But no behavioral >>>>> models so one cannot simulate in a reasonable time frame. >>>> Yet you will only use LTC because they're current mode. <boggle> >>>> >>> You are not correctly reading what I write. I am using LTC because there >>> are models that allow fast simulation. _Not_ because they are current >>> mode. I only require that any switcher that replaces the old one is >>> current mode. >> >> That's not what you said. You said it like current mode was something unique. > > >Where?"Some don't have current-mode control and that's a no-no here." Almost all do today. If you weren't making some point about it being somehow unique, it was a null statement.>> It surely isn't. >> >>>>>>>>> So far pretty >>>>>>>>> much all my designs came out almost verbatim as simulated. That's the >>>>>>>>> kind of time-to-market clients want. >>>>>>>> You trust behaviorals? Amazing. >>>>>>> Depends on who made them. From Linear Technology, yes, usually I trust >>>>>>> them. Many years of successful design with their models gives me confidence. >>>>>> I sure don't. I've never seen a behavioral that's good outside it's *NORMAL* >>>>>> operating conditions. That's never where the demons are. >>>>>> >>>>> Then you haven't tried hard enough with LTSpice :-) >>>> Nonsense. >>>> >>>>> By now they know me at LTC for lots of unorthodox uses. >>>> The models are useless outside their limited operational area. IOW, useless >>>> for guaranteeing operation. >>>> >>> Sorry, but I have proven that statement wrong many times. For that to >>> work you do need to establish good connections to their engineers >>> though. I have. >> >> YOu *think* you've simulated everything that can go wrong. Obviously a model >> can't be used outside it's limits. Behaviorals have very limited useful area >> and they're rarely documented. >> > >The same goes for full models. There are limits to everything virtual.Sure, but they're almost always more general than a behavioral. Most behaviorals don't even model their own supply current. Very few actually tell you *what* they're modeling.>However, with behavioral models I've had years of successful designs.I've had years of successful designs with little simulation at all.>And yeah, I did discover the occasional model bug and also the >occasional real bug on an IC. Where sim and real life did not jibe. >Usually it meant that the datasheet didn't either. That's part of life.Irrelevant. I'm not talking about bugs.
Reply by ●October 24, 20122012-10-24
On 10/23/2012 10:37 PM, krw@att.bizzzzzzzzzzzz wrote:> On Tue, 23 Oct 2012 21:05:00 -0400, rickman<gnuarm@gmail.com> wrote: > >> On 10/23/2012 8:30 PM, krw@att.bizzzzzzzzzzzz wrote: >>> On Tue, 23 Oct 2012 18:26:46 -0400, rickman<gnuarm@gmail.com> wrote: >>> >>>> On 10/22/2012 10:46 PM, krw@att.bizzzzzzzzzzzz wrote: >>>>> On Mon, 22 Oct 2012 18:02:03 -0700, Joerg<invalid@invalid.invalid> wrote: >>>>>> >>>>>> Well, I can't spend time bench-testing a new switcher chip that doesn't >>>>>> have a decent (meaning behavioral) SPICE model. My stuff usually goes >>>>>> CAD to production these days, with no lab bench time before the >>>>>> prototypes are done. On this one I only had to bench test the load and >>>>>> almost got sick when I looked at the results. >>>>> >>>>> You think you're alone? Get the manufacturer to guarantee the design. >>>> >>>> What good is that? So the board doesn't work like the simulation. Is >>>> the manufacturer going to fix the chip? They might fix the model and >>>> you still miss your schedule. >>> >>> They pay. Big time. Model? You're nuts. >> >> Hah! Pay!!! LOL!!! > > You are an idiot. > >> Good luck getting a chip maker to guarantee YOUR design no matter what >> you do. > > You're batting 1000 today. Go back to your g5eometry lessons, with Bloggs.I'm waiting for Joerg to get to this point with you. You get to where you can't reply so you start insulting people. I don't get why you bother. rick
Reply by ●October 24, 20122012-10-24
On Wed, 24 Oct 2012 20:18:20 -0400, rickman <gnuarm@gmail.com> wrote:>On 10/23/2012 10:37 PM, krw@att.bizzzzzzzzzzzz wrote: >> On Tue, 23 Oct 2012 21:05:00 -0400, rickman<gnuarm@gmail.com> wrote: >> >>> On 10/23/2012 8:30 PM, krw@att.bizzzzzzzzzzzz wrote: >>>> On Tue, 23 Oct 2012 18:26:46 -0400, rickman<gnuarm@gmail.com> wrote: >>>> >>>>> On 10/22/2012 10:46 PM, krw@att.bizzzzzzzzzzzz wrote: >>>>>> On Mon, 22 Oct 2012 18:02:03 -0700, Joerg<invalid@invalid.invalid> wrote: >>>>>>> >>>>>>> Well, I can't spend time bench-testing a new switcher chip that doesn't >>>>>>> have a decent (meaning behavioral) SPICE model. My stuff usually goes >>>>>>> CAD to production these days, with no lab bench time before the >>>>>>> prototypes are done. On this one I only had to bench test the load and >>>>>>> almost got sick when I looked at the results. >>>>>> >>>>>> You think you're alone? Get the manufacturer to guarantee the design. >>>>> >>>>> What good is that? So the board doesn't work like the simulation. Is >>>>> the manufacturer going to fix the chip? They might fix the model and >>>>> you still miss your schedule. >>>> >>>> They pay. Big time. Model? You're nuts. >>> >>> Hah! Pay!!! LOL!!! >> >> You are an idiot. >> >>> Good luck getting a chip maker to guarantee YOUR design no matter what >>> you do. >> >> You're batting 1000 today. Go back to your g5eometry lessons, with Bloggs. > >I'm waiting for Joerg to get to this point with you. You get to where >you can't reply so you start insulting people. I don't get why you >bother.Joerg has a brain and is a nice guy, unlike you, who are neither.
Reply by ●October 25, 20122012-10-25
krw@att.bizzzzzzzzzzzz wrote:> On Wed, 24 Oct 2012 10:33:46 -0700, Joerg <invalid@invalid.invalid> wrote: > >> krw@att.bizzzzzzzzzzzz wrote: >>> On Wed, 24 Oct 2012 07:58:18 -0700, Joerg <invalid@invalid.invalid> wrote: >>> >>>> krw@att.bizzzzzzzzzzzz wrote: >>>>> On Tue, 23 Oct 2012 17:49:19 -0700, Joerg <invalid@invalid.invalid> wrote: >>>>> >>>>>> krw@att.bizzzzzzzzzzzz wrote: >>>>>>> On Tue, 23 Oct 2012 14:19:16 -0700, Joerg <invalid@invalid.invalid> wrote: >>>>>>> >>>>>>>> krw@att.bizzzzzzzzzzzz wrote: >>>>>>>>> On Tue, 23 Oct 2012 12:45:14 -0700, Joerg <invalid@invalid.invalid> wrote: >>>>>>>>> >>>>>>>>>> krw@att.bizzzzzzzzzzzz wrote: >>>> [...] >>>> >>>>>> I do use others, just not for really critical stuff like this where >>>>>> simulation is key. >>>>> With a behavioral, you can only simulate what the model builder anticipated. >>>>> Again, that's rarely where the demons be. At least with full models, you have >>>>> a chance. >>>>> >>>> Yep. Last time I did that one single run too about four (!) hours. >>> <shrug> I've used entire mainframes for weeks on a single simulation. CPUs >>> weren't as cheap, nor as powerful, as they are now. Not everyone had one >>> sitting on their desk, either. >>> >> Well, I do not have a mainframe available here. > > You have about 100x the CPU power there. >Probably even more. But that kind of power is no accessible to everyone. I am hoping that cloud deals arise some day where you can "rent" CPU time at more reasonable deals. That would really be useful to folks who have to run tough sims.>>>>>>>> For some of my clients this is relevant. I have used PWM chips in the >>>>>>>> past that are 2nd-sourced. From Richtek and TI, for example. But those >>>>>>>> tend to be older designs. >>>>>>> Jellybeans, perhaps. Nothing leading edge. >>>>>>> >>>>>>>>>> Some >>>>>>>>>> don't have current-mode control and that's a no-no here. >>>>>>>>> Almost all do. Everything I've used is current-mode. >>>>>>>>> >>>>>>>> I am replacing a TI regulator that isn't, and this is one of the reasons. >>>>>>> Are you saying5 that TI doesn't have current mode regulators? <boggle> Almost >>>>>>> everything I've looked at in the last year is current mode. >>>>>>> >>>>>> Huh? That's not what I said. I said I am replacing a TI regulator that >>>>>> is not current-mode. It has to go because that one isn't up to the job. >>>>>> Of course they have developed better ones by now. But no behavioral >>>>>> models so one cannot simulate in a reasonable time frame. >>>>> Yet you will only use LTC because they're current mode. <boggle> >>>>> >>>> You are not correctly reading what I write. I am using LTC because there >>>> are models that allow fast simulation. _Not_ because they are current >>>> mode. I only require that any switcher that replaces the old one is >>>> current mode. >>> That's not what you said. You said it like current mode was something unique. >> >> Where? > > "Some don't have current-mode control and that's a no-no here."^^^^> > Almost all do today. If you weren't making some point about it being somehow > unique, it was a null statement. >Makes no sense to discuss this point further.>>> It surely isn't. >>> >>>>>>>>>> So far pretty >>>>>>>>>> much all my designs came out almost verbatim as simulated. That's the >>>>>>>>>> kind of time-to-market clients want. >>>>>>>>> You trust behaviorals? Amazing. >>>>>>>> Depends on who made them. From Linear Technology, yes, usually I trust >>>>>>>> them. Many years of successful design with their models gives me confidence. >>>>>>> I sure don't. I've never seen a behavioral that's good outside it's *NORMAL* >>>>>>> operating conditions. That's never where the demons are. >>>>>>> >>>>>> Then you haven't tried hard enough with LTSpice :-) >>>>> Nonsense. >>>>> >>>>>> By now they know me at LTC for lots of unorthodox uses. >>>>> The models are useless outside their limited operational area. IOW, useless >>>>> for guaranteeing operation. >>>>> >>>> Sorry, but I have proven that statement wrong many times. For that to >>>> work you do need to establish good connections to their engineers >>>> though. I have. >>> YOu *think* you've simulated everything that can go wrong. Obviously a model >>> can't be used outside it's limits. Behaviorals have very limited useful area >>> and they're rarely documented. >>> >> The same goes for full models. There are limits to everything virtual. > > Sure, but they're almost always more general than a behavioral. Most > behaviorals don't even model their own supply current. ...LTC's usually do.> ... Very few actually tell you *what* they're modeling. >True. Although if you have a good relationship with a manufacturer's engineers they'll help you there. In my case this has enabled several unorthodox designs and they were, consequently, rewarded with increased business driven their way.>> However, with behavioral models I've had years of successful designs. > > I've had years of successful designs with little simulation at all. >Our grandparents drove around without safety belts and survived :-)>> And yeah, I did discover the occasional model bug and also the >> occasional real bug on an IC. Where sim and real life did not jibe. >> Usually it meant that the datasheet didn't either. That's part of life. > > Irrelevant. I'm not talking about bugs.If you know the limitations of the models I find them extremely useful. I do miss the good old bench time though, because nowadays there's a lot less of that than 20 years ago. I have relied on simulators my whole career but of course in the 80's the computers weren't up to snuff for the real heavy lifting. -- Regards, Joerg http://www.analogconsultants.com/
Reply by ●October 25, 20122012-10-25
On 10/25/2012 1:14 PM, Joerg wrote:> krw@att.bizzzzzzzzzzzz wrote: > > Makes no sense to discuss this point further.You stopped a bit later than useful... ;^)>> I've had years of successful designs with little simulation at all. >> > > Our grandparents drove around without safety belts and survived :-)Uh, chances are that if your parents didn't have any children you won't either. Rick
Reply by ●October 30, 20122012-10-30
On 10/23/2012 12:40 PM, Joerg wrote:> miso wrote: >> On 10/22/2012 3:29 PM, Joerg wrote: >>> miso wrote: >>>> On 10/22/2012 2:22 PM, Joerg wrote: >>>>> Folks, >>>>> >>>>> Looked at Digikey and some others. Where are those huge capacitance >>>>> varicaps? The ones with several hundred pF of range for AM radios. All >>>>> gone lalaland by now? >>>>> >>>>> Even the ones I found in the 100pF range are either obsolete or not >>>>> recommended for new designs. >>>>> >>>>> What I am trying to do: I need to control a switcher chip in frequency >>>>> because I've got a very resonant load to deal with. Unfortunately it >>>>> sets the frequency with a timing cap. I'll have to somehow vary that >>>>> between 750pF and 2000pF. Can also be digital but then with a >>>>> granularity of 5pF. The sawtooth voltage across it is 2.5Vpp. >>>>> >>>>> Doing it with caps and a mux chip or two has its own challenges. The >>>>> ADG-series from AD is around 11pF per pin, otherwise their Rdson is too >>>>> high. Talking about the a rock and a hard spot here. >>>>> >>>>> Oh, and cost is not very important. If a diode or two or three are >>>>> needed that cost $5 a pop that's ok. >>>>> >>>> >>>> There are plenty of hoarders that have those caps. I picked them up >>>> years ago when I figured they would go extinct. They same goes for 10 >>>> turn pots of the panel variety with the dials on them. (Good old Mike >>>> Quinn in San Leadro by the Oakland airport would save panels with those >>>> pots and junk the rest of the product.] Most radios are synthesized >>>> these days, so there is no market for those caps. >>>> >>> >>> Yeah, I know but I was hoping there'd be at least one other market. >>> Seems like there ain't. >>> >>> >>>> If you are not conflating sawtooth with triangle wave, then maybe you >>>> could shunt some current from the osc pin, which would slow down the >>>> frequency. Pick your external cap for the highest frequency needed, then >>>> bleed with some DAC based circuit or even a resistive trimpot. The >>>> 10-turn PCB mounted resistors still exist. >>>> >>>> The difficulty will be in the compliance of the current source used for >>>> bleeding. >>> >>> >>> That won't be a problem but the chip immediately gets sea-sick when you >>> do anything DC to that pin. >>> >> >> Is this pin where you see the sawtooth? Seems to me with a high >> impedance current source, you can steal the current. Now a triangle >> waveform is a different case. >> >> So you have tried a high impedance current source? >> > > Yup. The chip then starts to misfire, big time, at least on the simulator. >None I ever designed would care. BTW, you can always make a sawtooth generator go faster by inserting a resistor in series with the cap. Basically it adds a step to the saw tooth, getting you to the limit faster.
Reply by ●October 31, 20122012-10-31
miso wrote:> On 10/23/2012 12:40 PM, Joerg wrote: >> miso wrote: >>> On 10/22/2012 3:29 PM, Joerg wrote: >>>> miso wrote: >>>>> On 10/22/2012 2:22 PM, Joerg wrote: >>>>>> Folks, >>>>>> >>>>>> Looked at Digikey and some others. Where are those huge capacitance >>>>>> varicaps? The ones with several hundred pF of range for AM radios. >>>>>> All >>>>>> gone lalaland by now? >>>>>> >>>>>> Even the ones I found in the 100pF range are either obsolete or not >>>>>> recommended for new designs. >>>>>> >>>>>> What I am trying to do: I need to control a switcher chip in >>>>>> frequency >>>>>> because I've got a very resonant load to deal with. Unfortunately it >>>>>> sets the frequency with a timing cap. I'll have to somehow vary that >>>>>> between 750pF and 2000pF. Can also be digital but then with a >>>>>> granularity of 5pF. The sawtooth voltage across it is 2.5Vpp. >>>>>> >>>>>> Doing it with caps and a mux chip or two has its own challenges. The >>>>>> ADG-series from AD is around 11pF per pin, otherwise their Rdson >>>>>> is too >>>>>> high. Talking about the a rock and a hard spot here. >>>>>> >>>>>> Oh, and cost is not very important. If a diode or two or three are >>>>>> needed that cost $5 a pop that's ok. >>>>>> >>>>> >>>>> There are plenty of hoarders that have those caps. I picked them up >>>>> years ago when I figured they would go extinct. They same goes for 10 >>>>> turn pots of the panel variety with the dials on them. (Good old Mike >>>>> Quinn in San Leadro by the Oakland airport would save panels with >>>>> those >>>>> pots and junk the rest of the product.] Most radios are synthesized >>>>> these days, so there is no market for those caps. >>>>> >>>> >>>> Yeah, I know but I was hoping there'd be at least one other market. >>>> Seems like there ain't. >>>> >>>> >>>>> If you are not conflating sawtooth with triangle wave, then maybe you >>>>> could shunt some current from the osc pin, which would slow down the >>>>> frequency. Pick your external cap for the highest frequency needed, >>>>> then >>>>> bleed with some DAC based circuit or even a resistive trimpot. The >>>>> 10-turn PCB mounted resistors still exist. >>>>> >>>>> The difficulty will be in the compliance of the current source used >>>>> for >>>>> bleeding. >>>> >>>> >>>> That won't be a problem but the chip immediately gets sea-sick when you >>>> do anything DC to that pin. >>>> >>> >>> Is this pin where you see the sawtooth? Seems to me with a high >>> impedance current source, you can steal the current. Now a triangle >>> waveform is a different case. >>> >>> So you have tried a high impedance current source? >>> >> >> Yup. The chip then starts to misfire, big time, at least on the >> simulator. >> > > None I ever designed would care. >Probably this one doesn't either, it's the simulator model that does. On this design I absolutely need a simulator model because of the ugly load. That shuts out most vendors.> BTW, you can always make a sawtooth generator go faster by inserting a > resistor in series with the cap. Basically it adds a step to the saw > tooth, getting you to the limit faster. >Can't do that here. The chip has internal comparators and certain things need to happen at certain times on the ramp. So I'd rather keep that ramp linear. -- Regards, Joerg http://www.analogconsultants.com/
Reply by ●October 31, 20122012-10-31
Joerg wrote:> miso wrote: > >>On 10/23/2012 12:40 PM, Joerg wrote: >> >>>miso wrote: >>> >>>>On 10/22/2012 3:29 PM, Joerg wrote: >>>> >>>>>miso wrote: >>>>> >>>>>>On 10/22/2012 2:22 PM, Joerg wrote: >>>>>> >>>>>>>Folks, >>>>>>> >>>>>>>Looked at Digikey and some others. Where are those huge capacitance >>>>>>>varicaps? The ones with several hundred pF of range for AM radios. >>>>>>>All >>>>>>>gone lalaland by now? >>>>>>> >>>>>>>Even the ones I found in the 100pF range are either obsolete or not >>>>>>>recommended for new designs. >>>>>>> >>>>>>>What I am trying to do: I need to control a switcher chip in >>>>>>>frequency >>>>>>>because I've got a very resonant load to deal with. Unfortunately it >>>>>>>sets the frequency with a timing cap. I'll have to somehow vary that >>>>>>>between 750pF and 2000pF. Can also be digital but then with a >>>>>>>granularity of 5pF. The sawtooth voltage across it is 2.5Vpp. >>>>>>> >>>>>>>Doing it with caps and a mux chip or two has its own challenges. The >>>>>>>ADG-series from AD is around 11pF per pin, otherwise their Rdson >>>>>>>is too >>>>>>>high. Talking about the a rock and a hard spot here. >>>>>>> >>>>>>>Oh, and cost is not very important. If a diode or two or three are >>>>>>>needed that cost $5 a pop that's ok. >>>>>>> >>>>>> >>>>>>There are plenty of hoarders that have those caps. I picked them up >>>>>>years ago when I figured they would go extinct. They same goes for 10 >>>>>>turn pots of the panel variety with the dials on them. (Good old Mike >>>>>>Quinn in San Leadro by the Oakland airport would save panels with >>>>>>those >>>>>>pots and junk the rest of the product.] Most radios are synthesized >>>>>>these days, so there is no market for those caps. >>>>>> >>>>> >>>>>Yeah, I know but I was hoping there'd be at least one other market. >>>>>Seems like there ain't. >>>>> >>>>> >>>>> >>>>>>If you are not conflating sawtooth with triangle wave, then maybe you >>>>>>could shunt some current from the osc pin, which would slow down the >>>>>>frequency. Pick your external cap for the highest frequency needed, >>>>>>then >>>>>>bleed with some DAC based circuit or even a resistive trimpot. The >>>>>>10-turn PCB mounted resistors still exist. >>>>>> >>>>>>The difficulty will be in the compliance of the current source used >>>>>>for >>>>>>bleeding. >>>>> >>>>> >>>>>That won't be a problem but the chip immediately gets sea-sick when you >>>>>do anything DC to that pin. >>>>> >>>> >>>>Is this pin where you see the sawtooth? Seems to me with a high >>>>impedance current source, you can steal the current. Now a triangle >>>>waveform is a different case. >>>> >>>>So you have tried a high impedance current source? >>>> >>> >>>Yup. The chip then starts to misfire, big time, at least on the >>>simulator. >>> >> >>None I ever designed would care. >> > > > Probably this one doesn't either, it's the simulator model that does. On > this design I absolutely need a simulator model because of the ugly > load. That shuts out most vendors. > > > >>BTW, you can always make a sawtooth generator go faster by inserting a >>resistor in series with the cap. Basically it adds a step to the saw >>tooth, getting you to the limit faster. >> > > > Can't do that here. The chip has internal comparators and certain things > need to happen at certain times on the ramp. So I'd rather keep that > ramp linear. >Have you tried the op-amp trick to vary a fixed capacitor? Depending on which way you need to shift it, lead of Lag will depend if you do a invert or non-inverting input. This is done commonly using a LM324 in many circuits., +/- rail supply Lead circuit +------------+ | | REf in | | + | |\| | REF out | +--+|-\ | ___ | | >-+---+----------+-|___|+---------+---------+ +----------+-|+/ | | /| | --- Lead Cap | +--------++ --- | | + + | | |\| | | + +--+|-\ + ___ | .-. | >+-++-|___|++ | |<-+----+--+|+/ | | |/| Lead Gain (D)'-' | === GND (created by AACircuit v1.28.6 beta 04/19/05 www.tech-chat.de) This isn't exactly what you would use but is shows how you can bring in a cap with the lead gain. This will insert the "Lead Cap" via the lead gain and forward reference coming in. Something to look at any way. Jamie
