This RF article is interesting - interpreting the empirical rules of foreigners for microwave circuits
[Copy link]
Interpreting the Foreigners' Rules of Experience for Microwave Circuits
Disclaimer:
The article to be interpreted comes from a foreign paper. Some of the individual interpretation points are quite interesting, so you can ask for everyone's opinions. If necessary, you can start a separate article to discuss them later.
1.
Microwave hybrid integrated circuits (hybrid modules) are relatively delicate, such as SIP (system in package), MCM, COP and the micro-assembly parts of frequency conversion components. When debugging and checking the circuit, do not touch the gold wire or bare chip with your fingers, otherwise the gold wire will be broken, and in serious cases, the chip will be damaged.
Author's note: Our colleagues who often play with micro-assembly should know that semi-finished products after micro-assembly are generally placed in a dust-free environment. During debugging, it is best to wear an anti-static wrist strap, protective clothing, and a mask to prevent saliva from splashing onto the chip.
2.
The minimum noise figure of a FET varies linearly with frequency up to fmax. This rule of thumb comes from John, who also provided a reference (thanks!). The minimum noise figure varies quadratically with frequency up to fmax.
The minimum noise figure (NFmin) of a FET varies linearly with frequency up to Fmax. The minimum noise figure of a BJT varies quadratically with frequency up to Fmax.
Additional Notes for Readers: Therefore, based on the project cost and comprehensive effect, select BJT, JEFT, HBT, HEMT and MESFET that are suitable for the operating frequency and have acceptable gain and noise figure. When the operating frequency is below 6GHz, bipolar transistors are mostly used; when the operating frequency is above 6GHz, field effect transistors are mostly used. In addition, the cutoff frequency of the transistor is usually required to be greater than or equal to 2-3 times the operating frequency. Low noise amplifiers require a higher cutoff frequency.
3.
Under the same substrate dielectric material and metal process conditions (in other words, the environment and circuit board remain unchanged), the loss of the branch line coupler decreases with the square root of the frequency. One of the reasons is the existence of the qufu effect.
Author's note: I don't know what you feel after reading this sentence. Do you feel that you haven't expanded your thoughts after learning about the skin effect? We just know that there is a skin effect, so when designing microstrip lines, we consider the requirements for coating thickness at different frequencies. Of course, if you learn statistics, different circuits have different rules. Once you master this rule, you will have a corresponding reference when designing again. For example, for a branch line coupler, if you know that there is a skin effect formula, the skin depth formula is d=k*66.1/f^0.5(mm), and the main line of the branch line coupler can be approximately considered as a microstrip line, and its loss is mainly composed of dielectric loss and conductor loss. When working below 10GHz, the conductor loss of the microstrip line is much greater than the dielectric loss, and the conductor loss can be approximated as R0/2Z0. R0 is inversely proportional to the skin depth. As the frequency increases, the smaller the skin depth, the larger R0. Therefore, the microstrip line loss and skin depth change as the square root.
I think the foreigner's statement is conditional. Maybe the author did not take the frequency into consideration. If the frequency exceeds 10GHz, it will not change in a square root relationship with the frequency. Therefore, what the foreigner said is not necessarily reliable.
4.
A conductor with a depth of 5 times the qufu can minimize the loss of the microstrip line.
Author's note: Actually, it is close. Foreigners' words are not very rigorous.
5.
If you use 1/2 ounce or thicker copper, you don't have to worry about qufu depth problems, provided you don't operate below 200MHz. If you're operating at microwave frequencies, thicker than 1/2 ounce copper will not reduce losses because you're reaching maximum surface conductivity. But if you're working on the bias circuits of high-power, high-current solid-state amplifiers, then adding more copper can reduce losses because all the copper is used for conduction. A "skin depth problem" means that you're not reaching at least three (preferably five) skin depths, so you're not doing the best you can to reduce losses.
Author's note: It is still necessary to pay attention to the relationship between frequency, power, temperature and qufu depth.
6.
Electromagnetic energy like microwave radiation travels in free space at a speed of one foot per nanosecond. In a Teflon-insulated coaxial cable, it moves one foot in 1.5 nanoseconds. In a waveguide, the speed is a function of frequency, due to dispersion.
7.
The return loss of a circulator is almost equal to its isolation.
8.
An amplifier's third order intercept point is usually 10 dB above its 1dB compression point when measured at the output, which equates to 9dB higher when measured at the input. There are exceptions, and the latest phemt devices usually have a higher TOI than you'd expect, so if your amplifier starts to compress at about 20dBm (out), then the TOI might be about 30dBm.
9.
If the other amplifier is still operational (to some extent), the isolation resistors used in the quadrature couplers (e.g. Lange) at the output of the amplifier must be able to handle 25% of the total power. Otherwise, 10% of the total power of the deployed hybrid circuit (hybrid), or 5% of the total power of the amplifier, will fall on the MMIC amplifier chip.
10.
For a given switch arm design, a single-pole double-throw switch (SPDT) will have 6dB higher isolation than a similar single-pole single-throw switch (SPST), provided the “through” arm of the switch is connected to a matched load.
11. For a gold wire with a diameter of 1 mil, the parasitic inductance is approximately equal to its length. The inductance of a bonding wire with a length of 40 mil is approximately 1 nH.
12.
"Lumped Components" The characteristics of any structure cannot exceed 1/10 of the wavelength at its maximum operating frequency.
13.
The thickness of a microstrip board should not exceed 1/10 of the wavelength at its maximum operating frequency, otherwise it will fail.
14.
How do you know what the waveguide number means just by looking at it? The Wr number represents the width of the waveguide divided by 10 in mils.
Author's Note: National Standard of the People's Republic of China Hollow Metal Waveguide GB 11450-891 General 1.1 Terminology Terminology shall be in accordance with the provisions of GB2900.1 "Basic Terminology of Electrical Terminology". 1.2 Model Nomenclature The model of the waveguide included in this standard consists of the following parts: a. The letter code B of the waveguide; b. The letter code indicating the shape of the internal cross-section of the waveguide; J - ordinary rectangle; B - flat rectangle; Z - medium flat rectangle; F - square; Y - round. c. A number that characterizes the specifications of a specific waveguide. This number indicates the frequency characteristics of the waveguide and approximately represents the geometric mean frequency of the main mode frequency range in multiples of 100MHz. For example: BJ100 represents a 22.860mmX10.160mm ordinary rectangular waveguide with a main mode center frequency of about 10GHz.
15.
A good way to remember which is the E-side and which is the H-side in a rectangular waveguide is that when you bend it, the E-side is "easy to bend" and the H-side is "hard to bend".
16.
For silicon or silicon germanium, 110 degrees Celsius is the maximum junction temperature for reliable operation (MTTF=1000,000). In addition to silicon process LDMOS, it can operate up to 175 degrees Celsius for 800 years. GaAsFET (or HEMT) channel temperature should not exceed 150 ℃ for long-term reliable operation. For gallium nitride HEMT (GaN), 175 ℃ is a good rule of thumb for the maximum communication temperature.
17.
To eliminate spurious modes, the width of the package should generally not exceed half the wavelength in free space at the maximum operating frequency.
Author's note: What does the work of structural engineers have to do with RF designers? However, structural engineers may not know this, so RF designers are still required to push structural engineers to perform eigenmode simulation. The foreigner finally used the word "generally". Sometimes the cavity vibration phenomenon is not just an intrinsic characteristic. When the RF energy is high or the link gain is high (the cavity has no effective gain segment isolation), cavity vibration phenomena of various modes often occur.
18.
For microstrip and stripline curves, use a minimum radius of three line widths at x-band and below. At higher frequencies, use a minimum radius of five line widths. Even better, use an optimal bevel instead of a curve.
19.
The 10% to 90% rise time of the pulse signal, in nanoseconds, is approximately equal to 0.35 divided by the network bandwidth, in GHz.
20. If you are trying to use quarter-wavelength stubs to achieve an RF short, use low-impedance wire, or better yet, use fan-shaped stubs.
Author's comment: This sentence applies the quarter-wavelength microstrip line matching theory. According to Zin=Z0×[ZL+ Z0*j*tg(2π/λ×L)]/[Z0+ ZL*j*tg(2π/λ×L)], L=1/4λ, ZL=0 (the capacitor impedance is approximately equal to 0), we can get Zin=Z0×j*tg(1/2×λ)=infinity, so it cannot be passed for RF signals, and it can also be used as an impedance transformer Zin=Z0 squared/ZL.
21. Due to constructive interference, the individual return losses of two identical mismatches are 6 dB better than the individual return losses measured together with the worst-case observed return losses of the two mismatches.
Author's note: What is constructive interference? The received signal is the superposition of two radio waves with a frequency of f and a phase difference of θ. When θ is an integer multiple of 2π, the two radio waves are constructively superimposed and the received signal is enhanced. Therefore, the two radio waves are called constructive interference. To further explain, constructive interference can generally be written as exp(-ikx)+exp(-ikx), destructive interference can be written as exp(-ikx)+exp(-ikx+pi), and standing wave can be written as exp(-ikx)+exp(ikx). In physical terms, constructive interference can be considered as two identical waves propagating in the same direction.
22.
Two similar mismatches can cancel each other by placing them approximately 1/4 (or perhaps 3/4) wavelength apart. This rule is often used in PIN diode switch and limiter design. Note that the VSWR of a capacitive shunt only needs to be slightly less than 1/4 wavelength to cancel, while the VSWR of an inductive shunt mismatch needs to be slightly more than 1/4 wavelength.
23.
Want to remember the correct order of the Ku/K/Ka radar bands? K is the middle band (18-27GHz), while Ku-band is lower in frequency (K - 'below') and Ka-band is higher in frequency (K - 'above').
24.
If the chip attenuator is mounted with conductive glue on a circuit board such as duroid or FR-4, it can pass at least 1/16W of power; if mounted on metal with conductive epoxy, it can pass 1/4W; if mounted on a metal heat sink with solder, it can pass 1/2W.
25.
For a 10 dB attenuation (90% power dissipation), the input resistor is subject to 1/2 of the maximum input power. For a 20 dB attenuation (99% power dissipation), it is subject to 80% of the maximum input power. For higher attenuation values, the input resistor is subject to the full RF input power.
26.
When designing the waveguide slot, tell your mechanical engineer to make the slot along the H plane. If you make the slot along the E plane, you will be asking for trouble, because the waveguide needs to conduct the RF current from this joint, resulting in high loss and poor VSWR.
27.
For an n-way resistor power divider, the division efficiency is (1/n)^2. Compare this to a lossless power divider, which has a division efficiency of (1/n), and you'll see that resistor dividers are extremely inefficient (and get worse as more legs are added), but for some applications they offer an inexpensive, broadband solution.
28.
For an impedance matched amplifier, as long as its ratio of s21 to s12 is at least 20 dB lower, the impedance match it sees on one port will not affect the impedance match it provides on the other port. Example: You are designing a receiver where the mixer has a very bad match at the IF port, such as 3:1 VSWR, or -6 dB return loss. The mixer is followed by a GaAs HBT amplifier with an S21 (gain) of 23 dB and an S12 (reverse isolation) of -25 dB. You are in trouble because the signal is only -2 dB different going round trip through the amplifier, so your receiver output match will only be 2 dB better than a 3:1 matched mixer, or -8 dB return loss.
29.
The P1dB compression point of the mixer RF input is generally 6dB lower than the LO drive power. However, we have also seen some examples where the P1dB is 10dB to 0dB lower than the LO power, so it is recommended to refer to the mixer manual.
30.
The temperature coefficient of gain for a fixed gate biased MESFET or PHEMT amplifier is typically -0.007dB/°C/stage. Self-biased amplifiers have a much lower gain temperature coefficient (gain changes much less with temperature).
31.
Capacitor materials with the highest dielectric constants generally have the largest changes with temperature. When it comes to capacitor materials, the materials with the highest dielectric constants change the worst with temperature.
32.
If you forget to include image rejection in your receiver design, the noise figure may increase by 3dB. About 20dB of image rejection will eliminate almost all of the folding noise.
33.
When using a microwave power meter to measure high power, cascade your linear coaxial attenuators in the order of 3dB, 6dB... closest to the power source. This way, each attenuator generates the least heat. Also, be sure to carefully check the output power of each attenuator that is screwed to the power source!
34.
If your antenna and coax haven't been struck by lightning or physically damaged, then use a spectrum analyzer to check the output of your transmitter. Perhaps you are seeing harmonics from a bad transmitter rather than a bad coax or antenna.
35.
The group delay of a filter is almost proportional to its order and inversely proportional to the bandwidth of the filter (a filter with a smaller bandwidth has a larger group delay). This "inference" comes from Chip (a person's name), and we haven't had time to carefully examine it: the insertion loss at the edge of the passband is equal to the insertion loss at the center of the passband multiplied by the ratio of the group delay at the edge of the passband to the group delay at the center of the passband (that is, the insertion loss is proportional to the length of time the signal is in the filter).
36.
The effective dielectric constant of the coplanar waveguide (CPW) is actually the average of the dielectric constants of the substrate medium and the vacuum dielectric constant. For example, the Er of GaAs is 12.9, so the effective dielectric constant of the CPW is (12.9+1)/2=6.95.
37.
The noise figure of a mixer is approximately equal to its conversion loss, or perhaps slightly less: a mixer with a conversion loss of -6dB might have a noise figure of 5.5dB.
38.
The return loss of the three mixer ports should be measured at the recommended LO drive power, otherwise the results will be poor.
39.
For best LO-IF port isolation, be sure to tap the IF output from the RF balun, not the LO balun. This should give you 20dB better LO suppression.
40.
The beam width of an antenna of a specific area is proportional to the wavelength, so a 40 GHz signal can be focused to 1/4 of the beam width of a 10 GHz signal.
41.
The coupled port of a microstrip or stripline directional coupler is the one closest to the input port because it is a reverse coupler . The coupled port of a waveguide broadside directional coupler is the one closest to the output port because it is a forward coupler .
42.
This is a "for reference only" rule of thumb, as there is no supporting data: for a microstrip line with a finite ground plane, the width of the ground plane needs to be at least five times the dielectric thickness or five times the width of the microstrip line, whichever is greater.
43.
To calculate the group delay by measuring the S parameters, the frequency interval should be set to a phase difference of about 10 degrees between adjacent frequency points S21. If the difference is less than 10 degrees, the measurement result may easily jump, while if the difference is greater than 10 degrees, some real problems hidden in the flatness of the group delay may be ignored. If you want to ask how to calculate the frequency interval, I'm sorry to wait for us to come up with a formula...
44.
If you divide the figure of merit of the switching element by 10 (figure of merit FOM = (1/(2pi*Ron*Coff)), you get the highest operating frequency that the element can be used for as a switch. So a MESFET switch can operate up to about 26GHz, a PHEMT switch to about 40GHz, and a PIN diode switch to about 180GHz.
45.
When counting the number of squares to calculate the resistance of a zigzag resistor, each corner should be counted as 1/2 square.
46.
The isolation of a switch is usually limited by the isolation of the package. If you design a 60dB switch, you should think carefully about the package!
47.
The noise figure of a linear passive device is equal to its loss. Calculated in dB, NF = S21 (dB). A device with a loss of 1dB has a noise figure of 1dB. There are more factors to consider! The above statement is only true if the linear passive device is at room temperature. You are better off analyzing this problem using noise temperature.
48.
If your LNA or receiver has a 20dB gain, then the noise figure contributed by the subsequent stage will be very small (unless the noise figure of the subsequent stage is horribly large!)
49.
As long as the image suppression is 20dB, the image folding noise can be basically ignored. In the worst case, the image folding noise will cause the receiver noise figure to deteriorate by 3dB.
50.
The impedance can only be calculated using the "conventional" closed form equation when the minimum width of the stripline edge covered by metal is 5 times the line width.
51.
The acceptable operating limit for a rectangular waveguide is (approximately) between 125% and 189% of the lower cutoff frequency. So for a WR-90, with a cutoff frequency of 6.557 GHz, the acceptable operating band is 8.2 to 12.4 GHz.
52.
For a given frequency, waveguide has the lowest loss per unit length. Coax will have losses about 10 times higher (in dB). Transmission line losses on an MMIC (microstrip or coplanar waveguide) are about 10 times worse than coax, or 100 times worse than waveguide (but the transmission line lengths are really small!) Stripline will usually have slightly higher losses than coax, depending on its geometry.
53.
Whenever you bend a transmission line, to model the length of the transmission line you simply ignore the extra length added by the bend. We would say this is only an approximation and if the effective length of the line is critical to the success of your design, you'd better simulate it in Sonnet!
54.
If you use a radius greater than three times the line width, the impedance characteristics of the transmission line will be almost indistinguishable from a straight line segment.
55.
The impedance of a coax is not a strong function of the eccentricity of the center conductor. You can be 50% off and the impedance will only decrease by about 10%! Remember, impedance can only decrease when the center conductor is off center, it can never increase! When designing a coaxial structure, you can never be perfectly concentric. Therefore, always design a coaxial structure to have a 3-6% higher impedance and you will end up with a better match.
56.
The Wilkinson isolation is 6 dB better than the return loss of the source matched at its common port.
57.
The return loss at the Wilkinson's shunt port is no better than the return loss the Wilkinson sees at its sum port.
58.
The acceptable voltage drop for a pulsed mode power amplifier is 5%, which will drop the power by a similar amount (5%, or about a quarter dB). So for a PHEMT amplifier operating at 8 volts, the allowed voltage drop is 0.4 volts. Use this rule when calculating charge storage capacitors!
59.
For microstrip lines, you can (approximately) halve the metal losses by doubling the dielectric thickness.
60.
When laying down the top layer of a microstrip board, many of us do ground fill. The question is how close to the microstrip line, especially since the ground fill function is performed automatically. The answer is to keep it > 3W. This ensures minimal additional losses and impact on the line impedance.
61.
Different loss types have different models for their relationship to frequency. Metallic losses are proportional to the square root frequency. Dielectric losses are proportional to frequency. Dielectric conductors are constant over frequency.
Author's note: So there is a paragraph in a certain article that is not very rigorous, maybe foreigners made some supplements to it.
62. When considering transmission line losses due to dielectric conductivity, if the dielectric resistivity is greater than 10,000 Ohm-cm, then it is negligible! This rules out nearly all substrates except Silicon, which can range from 1 Ohm-cm (more lossy) to 10,000 Ohm-cm (very expensive float zone Silicon). PTFE is 1E18 Ohm-cm!
63.
Let's call this a suggested rule of thumb (your input is greatly appreciated!) A transmission line (coax, microstrip CPW, stripline but not waveguide) can be considered low loss if the loss per wavelength is less than 0.1 dB. Waveguide will typically be 10x better than this benchmark.
Author's note: It is too general to measure the effect of transmission line by the loss of each wavelength being less than 0.1dB. It is still related to frequency, material, dielectric thickness, surface smoothness, process, etc. Moreover, at a frequency of 10GHz, the loss of each wavelength is expected to be 0.5dB, and it is unlikely to achieve a loss of 0.1dB.
64.
For stripline and microstrip, the attenuation factor always decreases as the characteristic impedance decreases. It's almost proportional; if you can use a 25 ohm transmission line instead of a 50 ohm one, you can cut the losses nearly in half! This is a different result than coaxial cable, which has a sweet spot on the attenuation/impedance curve (77 ohms for air coax, 52 ohms for PTFE-filled).
Author's comment: This sentence is interesting. Can we deliberately make the microstrip line wider and then match it to the appropriate value at the terminal?
65.
The typical isolation you can expect for a two-channel receiver is about 25 dB. For a two-channel MMIC, expect no more than 30 dB. Author's note: This is specific to MMICs.
66.
This rule comes from Cheryl... If you don't want to worry about the metal cap of the module changing the impedance of the microstrip circuit you designed, make sure it is a minimum height of 5 times the substrate thickness and a minimum of 5 times the maximum line width, whichever is larger.
67.
When a solid-state amplifier is pulsed for 100 us or longer ("long" pulses), it reaches a quasi-steady-state junction or channel temperature, so for thermal and reliability analysis the condition can be considered the same as continuous wave. Under the same operating conditions, to get the reliability of pulsed operation, you need to operate with pulses of 10 us or less ("short" pulses).
68.
The only thing the HBT is good at is being cheap and having low phase noise for the VCO, while the shorter gate length pHEMT is better in all other aspects.
69.
The effect of surface roughness on microstrip lines is a gradual increase in attenuation due to conductor losses. If the RMS roughness is approximately one skin depth, the conductor attenuation (alpha-c) increases by 60%. If the surface roughness is much greater than one skin depth, the increase is 100% (twice the ideal loss).
70.
How many sections are needed in a Wilkinson power divider? Divide the center frequency by the lowest frequency to get the answer. Therefore, for 2-18 GHz, plan five sections. Author's note: Generally, one octave increases by one level. There are 2 GHz more than 2 octaves in 2-18 GHz, so 5 sections are needed. Is this conclusion of the foreigner based on experience or theoretical derivation? I will not follow his idea for the time being. I am more confident in following the octave idea.
71.
For a two-way Wilkinson power divider, the worst case power dissipated in the isolation resistor can be easily calculated from the return loss of the two legs (no harmonic balance required). Assuming the mismatch seen by the two legs is 2:1, 11% of the power will be reflected back, and in the worst case, all of the power is dissipated in the isolation resistor. A 2:1 VSWR is -9.54 dB return loss, and 10^(-9.54/10) is 11%. However, this only happens if the mismatch is 180 degrees out of phase (if they were in phase, the reflected power would return to the Wilkinson input). Taking this to an extreme, if the loads have 0 dB return loss and are 180 degrees out of phase (like an open and a short), then the entire incident power to the Wilkinson power divider will be dissipated in the isolation resistor (the Wilkinson input ports appear matched!)
| 
提升卡
变色卡
千斤顶
