Why does WiFi signal drop by only one bar while internet speed drops by half? The router and phone were secretly shifting, so I copied down the gear shifter

a few nights ago, while watching a game in the bedroom, the 4K live stream card turned into a slideshow. He pulled out his phone to test speed: 80 Mbps. I remember clearly that the same phone tested in the living room was 940. But what about the WiFi icon? Full-gauge drops from three to two, only one space is lost.

one bar of signal, the speed dropped by ninety percent—no matter how you calculate it, it doesn't add up. At first, I suspected the router was old, but when I got back to the living room, it was 940Mbps again—the machine was fine. That night, I dug up everything and found the answer hidden in a mechanism I had never paid attention to: the router and phone kept secretly shifting gears.

run a lap first: the signal numbers and speed don't match at all

my AX3000T living room TV cabinet and gigabit broadband. I installed a WiFi analyzer on my Android phone, walked from the living room all the way to the bedroom, and casually jotted down the dBm (how do I read this number? I in the about blind spot self-checking, which is much more accurate than checking the bars): living room - 44, dining room - 55, master bedroom - 71, kitchen - 74, bathroom - 79.

test the speed in front of this watch, it gets interesting. Living Room - 44 Runs 940 Mbps Full Broadband; Restaurant -55 still has over 600 left; By the master bedroom to 71, only a little over 200 yuan remained; Kitchen - 74, fluctuating around 150; Bathroom - 79, speed detection software often runs nonstop.

signal drops from -44 to -71, which translates to a phone's bars drop by just one or two bars—manufacturers have compressed the wide range from -40 to -90 into three or four bars, with a single bar packed with worlds apart. But what really made me suspicious was another point: the speed didn't drop steadily with the signal, but down a step. From -55 to -65, a slight drop is made; from -65 to -71, it was cut in half and halved again. A stepwise decline must have a barrier mechanism behind it.

the number of bars depends on hearing, while the speed depends on the clarity of the

start by breaking down a concept. The number of signal bars reflects whether the signal has crossed the reception threshold. In plain language: the router is shouting, can you hear it? If you can hear, you get a rating; if your voice is loud, you get more space.

But network speed doesn't look at this; network speed depends on the signal-to-noise ratio (SNR): how many dB is the signal higher than the noise? WiFi channel noise floor noise fluctuates between -95 and -100dBm year-round, and you can't avoid it—microwaves, Bluetooth, neighbors' WiFi, dozens of smart devices at home—all add fuel to the fire.

do the math at my home: noise level -97 to get the living room -44, signal-to-noise ratio 53dB, which is like someone reading to you in a quiet library; Master bedroom: -71, signal-to-noise ratio down to just 26dB, like someone whispering to you at the market—you can hear the noise, but putting it together is a struggle.

here, a note about dB. In my about transmit power in , I calculated dB: dB is the ratio of logarithms, and every 3 plus actually doubles it. So the signal-to-noise ratio dropped from 53 to 26—not by half, but down to about one-thousandth of the original. The grid numbers are still lying to you; the physical quantities have long since collapsed.

QAM gear chart: How many pieces of cargo can a single trip handle

wireless data transmission isn't a slow flow through a water pipe, but truckload after truck. The number of bits each "train" (called symbol) depends on the modulation method, which is the QAM commonly seen on parameter pages. Here's my gear shifter:

Why are you afraid of noise in high gears? Imagine drawing constellations on a piece of parchment paper. QPSK has only 4 points, each occupying a large area. Even with a shake, you can still identify which one. 1024QAM requires squeezing 1024 dots onto the same paper, leaving only a hair's width between dots. If the noise is slightly pushed, this point is mistaken for the neighboring one—mistaken means a bit error, and if the code is wrong, you have to retransmit; too many retransmissions cause the speed to collapse.

engineering experience is very neat: each step up requires about 6dB more signal-to-noise ratio. QPSK can run just over 10dB, 256QAM requires 30dB, and 1024QAM requires over 35dB. So I drive at 53dB in the living room with 1024QAM speed, and 26dB in the master bedroom, only qualified to meet the threshold of 64QAM—the load per ride drops from 10 bits to 6 bits, so it's no wonder the speed is cut in half.

this also fills up the negotiation speed issue. In my two articles on writing spatial flow and writing bandwidth I broke down the formula: negotiated rate = number of spatial streams× bandwidth × bits per symbol. The first two are the number of lanes and lane width, while QAM is how much cargo each vehicle carries. Multiplying these three is the origin of the numbers 1201 and 2402 on your phone (bandwidth, negotiation, actual measurement the difference between the three ledgers I've written about this in another article).

shifting is fully automatic, and you never get told

the most crucial step: the gear isn't fixed during assembly—it changes automatically in real time. The mechanism is called link adaptation, or MCS in jargon: both sides contract each other, the receiving party tracks the bit error rate, and if there are too many mistakes, the other party is notified to downshift, then it stays steady for a while before trying to upgrade back. This adjustment happens every few dozen milliseconds, faster than a heartbeat.

you can watch it change with your own eyes. Windows computer connected to WiFi, open the command line and type:

netsh wlan show interfaces

look at the "Reception Rate" line. Carrying the laptop from the living room to the bedroom, the number jumps from 2,400 to 1,800, 1,400, 800, 400—each jump means a shift shift. It was the first time I stared at this number and walked through the whole room, and only then did I realize that the previous complaint about "WiFi speed fluctuating high" wasn't a malfunction at all; it was the gear setting diligently adjusting the signal to find balance.

this also conveniently explains two old incidents. One is "poor signal but no connection": at the lowest QPSK or even BPSK, each car only has one or two bits. Slow as long as you can hear it, it can be delivered, so you can chat on WeChat in the bathroom at -79 yuan. Another issue is why all the benefits of beamforming are in the corner—it concentrates energy in one direction, essentially pushing up the signal-to-noise ratio at the edges by a few dB, and only at high settings can it be used. I tested this for a week both on and off, .

want to stay stable at a high level, there are only three things you can do

Once you understand the

principle, the operation becomes natural. Signal-to-noise ratio = signal - noise. To maintain a stable high range, either raise the signal or suppress the noise; there is no third way.

first item: raise the signal. Moving the router is the cheapest option. the five positions I tested it: moving one meter can reduce the signal-to-noise ratio by 10dB, which is like getting two levels for free. If that doesn't work, use Mesh sub-routers to fill the gaps—don't expect your phone to bear the load.

the second item: choose the right frequency band. At the same location, 5G signal numbers look worse than 2.4G, but the 5G band is clean and has a high signal-to-noise ratio, so the range you can stick to is actually higher. I walked around the house and measured the difference before and after the reversal point in the of 2.4G and 5G—the essence is the gear difference. Try to stick your phone and computer to 5G, saving 2.4G for smart devices.

third, noise suppression. If your neighbor's WiFi is squeezed onto the same channel, and both of you are noisy, how to choose the channel? Ive written about Xiaomi's special (similar to other brands): 2.4G is 1/6/11, 5G is the emptiest. Many of their own smart devices also raise the underlying noise, moving away from IoT private networks or visitor networks, and the mainnet immediately becomes quiet.

also poured a couple of cold water on the matter. The amplifier and repeater can't save the gear—it already receives a low-level bad signal, and no matter how hard you try to relay, it only amplifies the bad signal, without changing the signal-to-noise ratio at all. I the I tested with the amplifier and cut the speed in half for this reason. There's also WiFi7's 4096QAM, which does have 12-bit per car, but requires a signal-to-noise ratio above 40dB. It only works when you're close to the router, and after a wall, it drops back to the same treatment as WiFi6. I calculated the cost before upgrading, but you really don't need to pay for bandwidth within gigabits.

Summary table: dBm, gear, phone swap—check in this order

combine the five points and gear points in my home into a comprehensive table, and you can match them accordingly:

check the order well: First, don't look at the number of bars—use a tool to check the dBm. For Android, install a random WiFi analyzer; if iOS doesn't have this permission, borrow an Android device or check Netsh on your computer; Step two: check the negotiation speed to see if it jumps. Jumping means you're shifting gears normally; if you don't move on a low number, it might be another issue. Step three: move the position and switch frequency bands to increase the signal-to-noise ratio; Step four: Check if the Xin Road is crowded. If you've maxed out the speed and still can't get the speed, it's not too late to talk about replacing the device—and according to this table, most people should first move the router out of the weak current box before replacing their device.